M.Sc. Courses at TU Delft
(AE4317) Autonomous Flight of Micro Air Vehicles
This course covers the challenges and existing state-of-the-art methods for enabling autonomous flight of Micro Air Vehicles (MAVs), ranging from 20-gram flapping wings to 1 kg quad rotors. The emphasis is on computationally efficient, bio-inspired approaches to MAV autonomous flight. The theoretical knowledge will be applied by the students in the practical assignment, in which student groups program quad rotors in order to avoid obstacles in TU Delft’s Cyberzoo.
https://www.youtube.com/watch?v=R3LNWh3UQHk
(AE4350) Bio-Inspired Intelligence and Learning for Autonomous Applications
This course covers the following subjects:
- Introduction Bio-inspired Intelligence
- Introduction to Artificial Neural Networks
- Introduction to Reinforcement Learning
- Reinforcement Learning for Aerospace Control
- Evolutionary robotics
- Self-supervised learning
M.Sc. Thesis at TU Delft
The MAVLab guides Master students from several faculties during their Master’s thesis. Please get in contact with us if you would like to pursue a thesis project with us.
Have a look at our recent list of M.Sc. thesis subjects:
2022
Dupon, Fréderic
UWB Localisation: Distributed UWB inter-ranging for MAV swarms in large GNSS-denied environments Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Pfeiffer, S.U. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:b7070c31-9db1-4a0c-8605-fb871914501b,
title = {UWB Localisation: Distributed UWB inter-ranging for MAV swarms in large GNSS-denied environments},
author = {Fréderic Dupon},
url = {http://resolver.tudelft.nl/uuid:b7070c31-9db1-4a0c-8605-fb871914501b},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {The use of micro air vehicles (MAV) is becoming increasingly mainstream and with them their applications have become more demanding across the board. The application of MAV’s in large GNSS-denied environments often asks for a distributed and scalable localisation system with minimal reliance on static localisation hardware. In this research a distributed ultra-wideband (UWB) localisation system that takes advantage of the collaborative capabilities of a swarm of MAV’s has been developed and tested in both simulation and practice. Additionally, a modular UWB simulator has been developed which enables researchers to test UWB localisation schemes for a swarm of MAV’s. It has been found that when taking advantage of the UWB inter-agent ranging capabilities of a swarm of micro air vehicles, one can increase the coverage of an UWB setup in spaces with coverage-issues and conversely increase the accuracy of an existing UWB setup that has full UWB coverage.},
note = {de Croon, G.C.H.E. (mentor); Pfeiffer, S.U. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
The use of micro air vehicles (MAV) is becoming increasingly mainstream and with them their applications have become more demanding across the board. The application of MAV’s in large GNSS-denied environments often asks for a distributed and scalable localisation system with minimal reliance on static localisation hardware. In this research a distributed ultra-wideband (UWB) localisation system that takes advantage of the collaborative capabilities of a swarm of MAV’s has been developed and tested in both simulation and practice. Additionally, a modular UWB simulator has been developed which enables researchers to test UWB localisation schemes for a swarm of MAV’s. It has been found that when taking advantage of the UWB inter-agent ranging capabilities of a swarm of micro air vehicles, one can increase the coverage of an UWB setup in spaces with coverage-issues and conversely increase the accuracy of an existing UWB setup that has full UWB coverage.
Campolucci, Pietro
Model and Actuator Based Trajectory Tracking for Incremental Nonlinear Dynamic Inversion Controllers Masters Thesis
TU Delft Aerospace Engineering, 2022, (Smeur, E.J.J. (mentor); Mancinelli, A. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:41895fac-aa59-47db-9c01-5e2879460b57,
title = {Model and Actuator Based Trajectory Tracking for Incremental Nonlinear Dynamic Inversion Controllers},
author = {Pietro Campolucci},
url = {http://resolver.tudelft.nl/uuid:41895fac-aa59-47db-9c01-5e2879460b57},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {This paper proposes a control strategy based on incremental nonlinear dynamic inversion (INDI), meant for trajectory tracking purposes. The controller extends the conven- tional capabilities of INDI by including actuator dynamics in the inversion law and introducing a state dependent compensation term to reduce the effort of the error controller. A complementary filter is employed to reduce the degrading effect introduced by the filtering-induced delay in the feedback loop. Both simulated and real flight tests are conducted on a quadrotor configuration with artificially slowed down actuators and a drag plate mounted on top, to better observe the effect of actuator dynamics and state dependent dynamics in trajectory tracking accuracy. Simulations show that the combination of the two additional features increases tracking accuracy both in the short and long term response. It is also found that an overestimation of the state compensation term leads to instability, which makes the strategy not robust to model mismatch. Real flight tests, involving the tracking of a series of doublets on the pitch attitude and a lemniscate of Bernoulli, show that, as the complexity of the maneuver increases, the less the state compensation term effectively contributes to an improved tracking when the model is incomplete. On the other hand, trajectory tracking accuracy due to the consideration of actuator dynamics shows consistency and improvement respect to conventional INDI solutions.},
note = {Smeur, E.J.J. (mentor); Mancinelli, A. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
This paper proposes a control strategy based on incremental nonlinear dynamic inversion (INDI), meant for trajectory tracking purposes. The controller extends the conven- tional capabilities of INDI by including actuator dynamics in the inversion law and introducing a state dependent compensation term to reduce the effort of the error controller. A complementary filter is employed to reduce the degrading effect introduced by the filtering-induced delay in the feedback loop. Both simulated and real flight tests are conducted on a quadrotor configuration with artificially slowed down actuators and a drag plate mounted on top, to better observe the effect of actuator dynamics and state dependent dynamics in trajectory tracking accuracy. Simulations show that the combination of the two additional features increases tracking accuracy both in the short and long term response. It is also found that an overestimation of the state compensation term leads to instability, which makes the strategy not robust to model mismatch. Real flight tests, involving the tracking of a series of doublets on the pitch attitude and a lemniscate of Bernoulli, show that, as the complexity of the maneuver increases, the less the state compensation term effectively contributes to an improved tracking when the model is incomplete. On the other hand, trajectory tracking accuracy due to the consideration of actuator dynamics shows consistency and improvement respect to conventional INDI solutions.
Chueca, Alejandro Barberia
Onboard Drone Detection with Event Cameras Masters Thesis
TU Delft Aerospace Engineering, 2022, (Dupeyroux, J.J.G. (mentor); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:be142c0a-3475-4571-b9c5-9118d397c51a,
title = {Onboard Drone Detection with Event Cameras},
author = {Alejandro Barberia Chueca},
url = {http://resolver.tudelft.nl/uuid:be142c0a-3475-4571-b9c5-9118d397c51a},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {In an effort to develop a new relative sensing method for drone swarms, the suitability of event cameras is assessed for propeller detection. Benchmark tests were conducted for different propellers under different lighting and background conditions, varying the observation distance and spinning frequency. The different tests were evaluated on event count, frequency, and clustering, as these are the most characteristic properties of the propeller-generated signal. A propeller detection metric was derived as a fuzzy classifier to assess detectability. It was observed that the sensor employed is limiting the detection range due to low resolution, with a maximum detection range of 75 cm. While at low spinning frequencies it is possible to detect the propeller at such distance, for higher frequences (6000 to 8000 RPMs) the range decreases to 45 cm for the tests with highest blade to background contrast and two-blade propellers. It was observed that lower contrasts reduce the successful detections only to low frequencies, and three-blade propellers become completely indetectable due to the static overlap between the blades. Therefore, it is concluded that, at this stage of the technology, the use case of event cameras for relative sensing is constrained to close distances with high contrast.},
note = {Dupeyroux, J.J.G. (mentor); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
In an effort to develop a new relative sensing method for drone swarms, the suitability of event cameras is assessed for propeller detection. Benchmark tests were conducted for different propellers under different lighting and background conditions, varying the observation distance and spinning frequency. The different tests were evaluated on event count, frequency, and clustering, as these are the most characteristic properties of the propeller-generated signal. A propeller detection metric was derived as a fuzzy classifier to assess detectability. It was observed that the sensor employed is limiting the detection range due to low resolution, with a maximum detection range of 75 cm. While at low spinning frequencies it is possible to detect the propeller at such distance, for higher frequences (6000 to 8000 RPMs) the range decreases to 45 cm for the tests with highest blade to background contrast and two-blade propellers. It was observed that lower contrasts reduce the successful detections only to low frequencies, and three-blade propellers become completely indetectable due to the static overlap between the blades. Therefore, it is concluded that, at this stage of the technology, the use case of event cameras for relative sensing is constrained to close distances with high contrast.
Eggers, Yvonne
Intrinsic Plasticity for Robust Event-Based Optic Flow Estimation Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Dupeyroux, J.J.G. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:3ffa7f45-8631-4224-a16b-4e2be097e35b,
title = {Intrinsic Plasticity for Robust Event-Based Optic Flow Estimation},
author = {Yvonne Eggers},
url = {http://resolver.tudelft.nl/uuid:3ffa7f45-8631-4224-a16b-4e2be097e35b},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Event cameras and spiking neural networks (SNNs) allow for a highly bio-inspired, low-latency and power efficient implementation of optic flow estimation. Just recently, a hierarchical SNN was proposed in which motion selectivity is learned from raw event data in an unsupervised manner using spike-timing-dependent plasticity (STDP). However, real-life applications of this SNN are currently still limited by the fact that the exact choice of neuron parameters depends on the spatiotemporal properties of the input. Furthermore, tuning the network is a challenging task due to the high degree of coupling between the various parameters. Inspired by neurons in biological brains that modify their intrinsic parameters through a process called intrinsic plasticity, this research proposes update rules which adapt the voltage threshold and maximum synaptic delay during inference. This allows applying the already trained network to a wider range of operating conditions and simplifies the tuning process. Starting with a detailed parameter analysis, primary functions and undesired side effects are assigned to each parameter. The update rules are then designed in such a way as to eliminate these side effects. Unlike existing update rules for the voltage threshold, this work does not attempt to keep the firing activity of output neurons within a specific range, but instead aims to adjust the threshold such that only the correct output maps spike. In particular, the voltage threshold is adapted such that output spikes occur in no more than two maps per retinotopic location. The maximum synaptic delay is adapted such that the resulting apparent pixel velocities of the input match those of the data used during training. A sensitivity analysis is presented which illustrates the effects of newly introduced parameters on the network performance. Furthermore, the adapted network is tested on real event data recorded onboard a drone avoiding obstacles. Due to the difficulties in matching the output of the adapted SNN to the ground truth data, quantitative results are inconclusive. However, qualitative results show a clear improvement in both the density and correctness of optic flow estimates.},
note = {de Croon, G.C.H.E. (mentor); Dupeyroux, J.J.G. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Event cameras and spiking neural networks (SNNs) allow for a highly bio-inspired, low-latency and power efficient implementation of optic flow estimation. Just recently, a hierarchical SNN was proposed in which motion selectivity is learned from raw event data in an unsupervised manner using spike-timing-dependent plasticity (STDP). However, real-life applications of this SNN are currently still limited by the fact that the exact choice of neuron parameters depends on the spatiotemporal properties of the input. Furthermore, tuning the network is a challenging task due to the high degree of coupling between the various parameters. Inspired by neurons in biological brains that modify their intrinsic parameters through a process called intrinsic plasticity, this research proposes update rules which adapt the voltage threshold and maximum synaptic delay during inference. This allows applying the already trained network to a wider range of operating conditions and simplifies the tuning process. Starting with a detailed parameter analysis, primary functions and undesired side effects are assigned to each parameter. The update rules are then designed in such a way as to eliminate these side effects. Unlike existing update rules for the voltage threshold, this work does not attempt to keep the firing activity of output neurons within a specific range, but instead aims to adjust the threshold such that only the correct output maps spike. In particular, the voltage threshold is adapted such that output spikes occur in no more than two maps per retinotopic location. The maximum synaptic delay is adapted such that the resulting apparent pixel velocities of the input match those of the data used during training. A sensitivity analysis is presented which illustrates the effects of newly introduced parameters on the network performance. Furthermore, the adapted network is tested on real event data recorded onboard a drone avoiding obstacles. Due to the difficulties in matching the output of the adapted SNN to the ground truth data, quantitative results are inconclusive. However, qualitative results show a clear improvement in both the density and correctness of optic flow estimates.
Ferede, Robin
An Adaptive Control Strategy for Neural Network based Optimal Quadcopter Controllers Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Wagter, C. (mentor); de Croon, G.C.H.E. (mentor); Izzo, Dario (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:b43a9703-082c-47c7-a56e-d50794ee8c1c,
title = {An Adaptive Control Strategy for Neural Network based Optimal Quadcopter Controllers},
author = {Robin Ferede},
url = {http://resolver.tudelft.nl/uuid:b43a9703-082c-47c7-a56e-d50794ee8c1c},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Developing optimal controllers for aggressive high speed quadcopter flight remains a major challenge in the field of robotics. Recent work has shown that neural networks trained with supervised learning are a good candidate for real-time optimal quadcopter control. In these methods, the networks (termed G&CNets) are trained using optimal trajectories obtained from a dynamical model of the quadcopter by means of a direct transcription method. A major problem with these methods is the effects of unmodeled dynamics. In this work we identify these effects for G&CNets trained for power optimal full state-to-rpm feedback. We propose an adaptive control strategy to mitigate the effects of unmodeled roll, pitch and yaw moments. Our method works by generating optimal trajectories with constant external moments added to the model and training a network to learn the policy that maps state and external moments to the corresponding optimal rpm command. We demonstrate the effectiveness of our method by performing power-optimal hover-to-hover flights with and without moment feedback. The flight tests show that the inclusion of this moment feedback significantly improves the controller's performance. Additionally we compare the adaptive controller's performance to a time optimal Bang-Bang controller for consecutive waypoint flight and show significantly faster lap times on a 3x4m track.},
note = {de Wagter, C. (mentor); de Croon, G.C.H.E. (mentor); Izzo, Dario (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Developing optimal controllers for aggressive high speed quadcopter flight remains a major challenge in the field of robotics. Recent work has shown that neural networks trained with supervised learning are a good candidate for real-time optimal quadcopter control. In these methods, the networks (termed G&CNets) are trained using optimal trajectories obtained from a dynamical model of the quadcopter by means of a direct transcription method. A major problem with these methods is the effects of unmodeled dynamics. In this work we identify these effects for G&CNets trained for power optimal full state-to-rpm feedback. We propose an adaptive control strategy to mitigate the effects of unmodeled roll, pitch and yaw moments. Our method works by generating optimal trajectories with constant external moments added to the model and training a network to learn the policy that maps state and external moments to the corresponding optimal rpm command. We demonstrate the effectiveness of our method by performing power-optimal hover-to-hover flights with and without moment feedback. The flight tests show that the inclusion of this moment feedback significantly improves the controller's performance. Additionally we compare the adaptive controller's performance to a time optimal Bang-Bang controller for consecutive waypoint flight and show significantly faster lap times on a 3x4m track.
Verheyen, Jan
Insect-Inspired Visual Guidance: are current familiarity-based models ready for long-ranged navigation? Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Dupeyroux, J.J.G. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:823d959a-17b8-4fd9-bc45-a0ace45d29ca,
title = {Insect-Inspired Visual Guidance: are current familiarity-based models ready for long-ranged navigation?},
author = {Jan Verheyen},
url = {http://resolver.tudelft.nl/uuid:823d959a-17b8-4fd9-bc45-a0ace45d29ca},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Insects have — over millions of years of evolution — perfected many of the systems that roboticists aim to achieve; they can swiftly and robustly navigate through different environments under various conditions while at the same time being highly energy efficient. To reach this level of performance and efficiency one might want to look at and take inspiration from how these insects achieve their feats. Currently, no dataset exists that allows bio-inspired navigation models to be evaluated over long real- life routes. We present a novel dataset containing omnidirectional event vision, frame-based vision, depth frames, inertial measurement (IMU) readings, and centimeter-accurate GNSS positioning over kilometer long stretches in and around the TUDelft campus. The dataset is used to evaluate familiarity-based insect-inspired neural navigation models on their performance over longer sequences. It demonstrates that current scene familiarity models are not suited for long-ranged navigation, at least not in their current form.},
note = {de Croon, G.C.H.E. (mentor); Dupeyroux, J.J.G. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Insects have — over millions of years of evolution — perfected many of the systems that roboticists aim to achieve; they can swiftly and robustly navigate through different environments under various conditions while at the same time being highly energy efficient. To reach this level of performance and efficiency one might want to look at and take inspiration from how these insects achieve their feats. Currently, no dataset exists that allows bio-inspired navigation models to be evaluated over long real- life routes. We present a novel dataset containing omnidirectional event vision, frame-based vision, depth frames, inertial measurement (IMU) readings, and centimeter-accurate GNSS positioning over kilometer long stretches in and around the TUDelft campus. The dataset is used to evaluate familiarity-based insect-inspired neural navigation models on their performance over longer sequences. It demonstrates that current scene familiarity models are not suited for long-ranged navigation, at least not in their current form.
Oever, Erik
An artificial neural network based method for grid-free acoustic source localization using multiple input frequencies Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:a5713055-c4a4-4a6e-8cdc-4c2ac1e4e300,
title = {An artificial neural network based method for grid-free acoustic source localization using multiple input frequencies},
author = {Erik Oever},
url = {http://resolver.tudelft.nl/uuid:a5713055-c4a4-4a6e-8cdc-4c2ac1e4e300},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {In recent years, efforts are focused on developing an acoustic based autonomous detect and avoidance system for UAVs to minimize interference with other air traffic. The purpose of this research is to study the potential of artificial neural networks for fast, grid-free acoustic source localization. A multi-layer perceptron has been trained to localize simulated white noise acoustic point sources using a converted version of the cross spectral matrix. The ANN based method shows similar localization behaviour to different frequencies as conventional beamforming. A new ANN architecture is proposed that uses the converted cross spectral matrices of multiple different frequencies as input to improve the localization accuracy. The multi input model has shown to have a mean absolute error of approximately 0.27[m]. The proposed model has also been applied on real world recording data of an aircraft flyover. The ANN based method has shown to be able to obtain a prediction within approximately 0.05[s], compared to approximately 1000-2000[s] for conventional beamforming. However, the magnitude and inconsistency of the localization error for the recording is higher compared to the simulated white noise source.},
note = {de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
In recent years, efforts are focused on developing an acoustic based autonomous detect and avoidance system for UAVs to minimize interference with other air traffic. The purpose of this research is to study the potential of artificial neural networks for fast, grid-free acoustic source localization. A multi-layer perceptron has been trained to localize simulated white noise acoustic point sources using a converted version of the cross spectral matrix. The ANN based method shows similar localization behaviour to different frequencies as conventional beamforming. A new ANN architecture is proposed that uses the converted cross spectral matrices of multiple different frequencies as input to improve the localization accuracy. The multi input model has shown to have a mean absolute error of approximately 0.27[m]. The proposed model has also been applied on real world recording data of an aircraft flyover. The ANN based method has shown to be able to obtain a prediction within approximately 0.05[s], compared to approximately 1000-2000[s] for conventional beamforming. However, the magnitude and inconsistency of the localization error for the recording is higher compared to the simulated white noise source.
Schröter, Shawn
We fly as one: Design and Joint Control of a Conjoined Biplane and Quadrotor Masters Thesis
TU Delft Aerospace Engineering, 2022, (Smeur, E.J.J. (mentor); Remes, B.D.W. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:703d5b28-75c1-4d8b-a1a6-93510aed7b29,
title = {We fly as one: Design and Joint Control of a Conjoined Biplane and Quadrotor},
author = {Shawn Schröter},
url = {http://resolver.tudelft.nl/uuid:703d5b28-75c1-4d8b-a1a6-93510aed7b29},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Unmanned Aerial Vehicles, UAVs, serve many purposes<br/>these days, such as short-range inspections<br/>and long-distance search and rescue missions. Long-distance missions can entail a search in a building. Such missions require a large aircraft for endurance and a small aircraft for manoeuvrability in a building.<br/><br/>This paper proposes a novel combination of a quadrotor and a hybrid biplane capable of joint hover, joint forward flight, and mid-air disassembly followed by separate flight. During joint flight, the quadcopter and the biplane have no intercommunication.<br/><br/>This paper covers the design of a release system and a joint control strategy. Firstly, the in-flight<br/>release is successfully tested in joint hover up to a forward pitch angle of -18 [deg]. Secondly, three control strategies for the quadrotor are compared:<br/>a proportional angular rate damper, a proportional angular acceleration damper, and constant thrust without attitude control.<br/>In all cases, the biplane uses a cascaded INDI attitude controller. Simulation and practical tests show that for intentional attitude changes, the different strategies<br/>are of minimal influence. However, the angular rate damper<br/>strategy for disturbance rejection has the lowest roll angle error and requires the smallest input command.<br},
note = {Smeur, E.J.J. (mentor); Remes, B.D.W. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Unmanned Aerial Vehicles, UAVs, serve many purposes<br/>these days, such as short-range inspections<br/>and long-distance search and rescue missions. Long-distance missions can entail a search in a building. Such missions require a large aircraft for endurance and a small aircraft for manoeuvrability in a building.<br/><br/>This paper proposes a novel combination of a quadrotor and a hybrid biplane capable of joint hover, joint forward flight, and mid-air disassembly followed by separate flight. During joint flight, the quadcopter and the biplane have no intercommunication.<br/><br/>This paper covers the design of a release system and a joint control strategy. Firstly, the in-flight<br/>release is successfully tested in joint hover up to a forward pitch angle of -18 [deg]. Secondly, three control strategies for the quadrotor are compared:<br/>a proportional angular rate damper, a proportional angular acceleration damper, and constant thrust without attitude control.<br/>In all cases, the biplane uses a cascaded INDI attitude controller. Simulation and practical tests show that for intentional attitude changes, the different strategies<br/>are of minimal influence. However, the angular rate damper<br/>strategy for disturbance rejection has the lowest roll angle error and requires the smallest input command.<br
LU, Jingyi
Evolving-to-Learn with Spiking Neural Networks Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Hagenaars, J.J. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:3e2b645f-5ef2-41f5-9e8f-70d64fc8b2a6,
title = {Evolving-to-Learn with Spiking Neural Networks},
author = {Jingyi LU},
url = {http://resolver.tudelft.nl/uuid:3e2b645f-5ef2-41f5-9e8f-70d64fc8b2a6},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Inspired by the natural nervous system, synaptic plasticity rules are applied to train spiking neural networks. Different from learning algorithms such as propagation and evolution that are widely used to train spiking neural networks, synaptic plasticity rules learn the parameters with local information, making them suitable for online learning on neuromorphic hardware. However, when such rules are implemented to learn different new tasks, they usually require a significant amount of work on task-dependent fine-tuning. This thesis aims to make this process easier by employing an evolutionary algorithm that evolves suitable synaptic plasticity rules for the task at hand. More specifically, we provide a set of various local signals, a set of mathematical operators, and a global reward signal, after which a Cartesian genetic programming process finds an optimal learning rule from these components. In this work, we first test the algorithm in basic binary pattern classification tasks. Then, using this approach, we find learning rules that successfully solve an XOR and cart-pole task, and discover new learning rules that outperform the baseline rules from literature.},
note = {de Croon, G.C.H.E. (mentor); Hagenaars, J.J. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Inspired by the natural nervous system, synaptic plasticity rules are applied to train spiking neural networks. Different from learning algorithms such as propagation and evolution that are widely used to train spiking neural networks, synaptic plasticity rules learn the parameters with local information, making them suitable for online learning on neuromorphic hardware. However, when such rules are implemented to learn different new tasks, they usually require a significant amount of work on task-dependent fine-tuning. This thesis aims to make this process easier by employing an evolutionary algorithm that evolves suitable synaptic plasticity rules for the task at hand. More specifically, we provide a set of various local signals, a set of mathematical operators, and a global reward signal, after which a Cartesian genetic programming process finds an optimal learning rule from these components. In this work, we first test the algorithm in basic binary pattern classification tasks. Then, using this approach, we find learning rules that successfully solve an XOR and cart-pole task, and discover new learning rules that outperform the baseline rules from literature.
Tran, Tommy
Semantic Segmentation using Deep Neural Networks for MAVs Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Xu, Y. (mentor); de Wagter, C. (graduation committee); van Gemert, J.C. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:7735d01c-b4cd-4173-a584-652f269c078c,
title = {Semantic Segmentation using Deep Neural Networks for MAVs},
author = {Tommy Tran},
url = {http://resolver.tudelft.nl/uuid:7735d01c-b4cd-4173-a584-652f269c078c},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Semantic segmentation methods have been developed and applied to single images for object segmentation. However, for robotic applications such as high-speed agile Micro Air Vehicles (MAVs) in Autonomous Drone Racing (ADR), it is more interesting to consider temporal information as video sequences are correlated over time. In this work, we evaluate the performance of state-of-the-art methods such as Recurrent Neural Networks (RNNs), 3D Convolutional Neural Networks (CNNs), and optical flow for video semantic segmentation in terms of accuracy and inference speed on three datasets with different camera motion configurations. The results show that using an RNN with convolutional operators outperforms all methods and achieves a performance boost of 10.8% on the KITTI (MOTS) dataset with 3 degrees of freedom (DoF) motion and a small 0.6% improvement on the CyberZoo dataset with 6 DoF motion over the single-frame-based semantic segmentation method. The inference speed was measured on the CyberZoo dataset, achieving 321 fps on an NVIDIA GeForce RTX 2060 GPU and 30 fps on an NVIDIA Jetson TX2 mobile computer.},
note = {de Croon, G.C.H.E. (mentor); Xu, Y. (mentor); de Wagter, C. (graduation committee); van Gemert, J.C. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Semantic segmentation methods have been developed and applied to single images for object segmentation. However, for robotic applications such as high-speed agile Micro Air Vehicles (MAVs) in Autonomous Drone Racing (ADR), it is more interesting to consider temporal information as video sequences are correlated over time. In this work, we evaluate the performance of state-of-the-art methods such as Recurrent Neural Networks (RNNs), 3D Convolutional Neural Networks (CNNs), and optical flow for video semantic segmentation in terms of accuracy and inference speed on three datasets with different camera motion configurations. The results show that using an RNN with convolutional operators outperforms all methods and achieves a performance boost of 10.8% on the KITTI (MOTS) dataset with 3 degrees of freedom (DoF) motion and a small 0.6% improvement on the CyberZoo dataset with 6 DoF motion over the single-frame-based semantic segmentation method. The inference speed was measured on the CyberZoo dataset, achieving 321 fps on an NVIDIA GeForce RTX 2060 GPU and 30 fps on an NVIDIA Jetson TX2 mobile computer.
Groen, Chris
Grammatical Evolution for Optimising Drone Behaviors Masters Thesis
TU Delft Aerospace Engineering, 2022, (Li, S. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:0fc90d7b-7aa3-4501-be7f-ac31330957b6,
title = {Grammatical Evolution for Optimising Drone Behaviors},
author = {Chris Groen},
url = {http://resolver.tudelft.nl/uuid:0fc90d7b-7aa3-4501-be7f-ac31330957b6},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {This paper reviews the application of grammatical evolution for the optimisation of low level parameters and high level behaviors for two drone behaviors, namely wall-following and navigation. In order to optimise these low level parameters and high level behaviors, grammatical evolution was applied to behavior trees. Grammatical evolution provided a significant improvement in the wall-following behavior of a drone, creating a more robust behavior. There was no improvement for the navigation behavior however, with the success rate of navigating deteriorating in some cases. The evolved wallfollowing behavior was compared and tested against another wall-following controller from literature, and shown to be superior. A real-life experiment was also conducted for the wall-following behavior, which led to positive results after correcting for the reality gap. For the wall-following behavior, the grammatical evolution promoted a continuous scanning behavior, which greatly increased it’s awareness of obstacles. Significant recommendations were given to improve the results of the grammatical evolution for both behaviors.},
note = {Li, S. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
This paper reviews the application of grammatical evolution for the optimisation of low level parameters and high level behaviors for two drone behaviors, namely wall-following and navigation. In order to optimise these low level parameters and high level behaviors, grammatical evolution was applied to behavior trees. Grammatical evolution provided a significant improvement in the wall-following behavior of a drone, creating a more robust behavior. There was no improvement for the navigation behavior however, with the success rate of navigating deteriorating in some cases. The evolved wallfollowing behavior was compared and tested against another wall-following controller from literature, and shown to be superior. A real-life experiment was also conducted for the wall-following behavior, which led to positive results after correcting for the reality gap. For the wall-following behavior, the grammatical evolution promoted a continuous scanning behavior, which greatly increased it’s awareness of obstacles. Significant recommendations were given to improve the results of the grammatical evolution for both behaviors.
Stikker, Roelof
Self-supervised finetuning of stereo matching algorithms Masters Thesis
TU Delft Aerospace Engineering, 2022, (de Croon, G.C.H.E. (mentor); Pfeiffer, S.U. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:a4305f40-c095-45c2-bc7e-84e23efa70d6,
title = {Self-supervised finetuning of stereo matching algorithms},
author = {Roelof Stikker},
url = {http://resolver.tudelft.nl/uuid:a4305f40-c095-45c2-bc7e-84e23efa70d6},
year = {2022},
date = {2022-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Abstract— Stereo vision is a commonly applied method to achieve depth perception on Micro Air Vehicles (MAVs). Stereo matching algorithms are often optimized for specific environments and camera properties, using the ground truth error as a supervisor. However, in practical applications ground truth data is usually not available. Therefore, in this research, we finetune several conventional stereo matching algorithms (BM, SGBM, and ELAS) and a neural network (AnyNet) using self-supervision. The settings of the conventional algorithms are optimized with NSGA-II, using the reconstruction error and disparity density as objective functions. AnyNet is finetuned with the reconstruction error, as well as with the disparity map of conventional methods. We conclude that finetuning the parameters of conventional stereo algorithms using the reconstruction error can lead to a slight improvement in performance compared with the general settings, depending on the stereo algorithm. The performance of the conventional methods is comparable to that of AnyNet on a major portion of the image. However, removing the values with low confidence in the disparity map of ELAS and interpolating the missing disparities leads to an accuracy well above AnyNet.},
note = {de Croon, G.C.H.E. (mentor); Pfeiffer, S.U. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Abstract— Stereo vision is a commonly applied method to achieve depth perception on Micro Air Vehicles (MAVs). Stereo matching algorithms are often optimized for specific environments and camera properties, using the ground truth error as a supervisor. However, in practical applications ground truth data is usually not available. Therefore, in this research, we finetune several conventional stereo matching algorithms (BM, SGBM, and ELAS) and a neural network (AnyNet) using self-supervision. The settings of the conventional algorithms are optimized with NSGA-II, using the reconstruction error and disparity density as objective functions. AnyNet is finetuned with the reconstruction error, as well as with the disparity map of conventional methods. We conclude that finetuning the parameters of conventional stereo algorithms using the reconstruction error can lead to a slight improvement in performance compared with the general settings, depending on the stereo algorithm. The performance of the conventional methods is comparable to that of AnyNet on a major portion of the image. However, removing the values with low confidence in the disparity map of ELAS and interpolating the missing disparities leads to an accuracy well above AnyNet.
2021
Beurden, Bas
Scalable Positioning Method for MAV Localisation using Two onboard UWB Tags Masters Thesis
TU Delft Aerospace Engineering, 2021, (Pfeiffer, S.U. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:bb6ace70-512e-4e0e-a834-6b065ece52c5,
title = {Scalable Positioning Method for MAV Localisation using Two onboard UWB Tags},
author = {Bas Beurden},
url = {http://resolver.tudelft.nl/uuid:bb6ace70-512e-4e0e-a834-6b065ece52c5},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Abstract—Ultra-wideband (UWB) ranging is a very suitable method for indoor localisation of unmanned aerial vehicles (UAVs). Current solutions of UWB ranging however either focus on achieving a high accuracy or focus on scalability. In this research a positioning algorithm for UAVs is presented that combines high accuracy performance with a high level of system scalability. The localisation method uses commercially available off the shelf components and is implemented by connecting two UWB sensors to a micro aerial vehicle. From<br/>both sensors, time-difference of arrival (TDOA) measurements were collected during flights and additionally, a tag-TDOA between the two UWB sensors was measured which estimates the angle-of-arrival of the incoming signals. It was found that state estimation using TDOA measurements from both UWB sensors has a reduced positioning error compared to the algorithm using TDOA measurements from one UWB sensor, without significantly affecting yaw estimation accuracy. Furthermore, the tag-TDOA measurement did not improve the estimation accuracy at the implemented baseline of 0.22 metres as the<br/>measurement error was too large compared to the baseline.},
note = {Pfeiffer, S.U. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Abstract—Ultra-wideband (UWB) ranging is a very suitable method for indoor localisation of unmanned aerial vehicles (UAVs). Current solutions of UWB ranging however either focus on achieving a high accuracy or focus on scalability. In this research a positioning algorithm for UAVs is presented that combines high accuracy performance with a high level of system scalability. The localisation method uses commercially available off the shelf components and is implemented by connecting two UWB sensors to a micro aerial vehicle. From<br/>both sensors, time-difference of arrival (TDOA) measurements were collected during flights and additionally, a tag-TDOA between the two UWB sensors was measured which estimates the angle-of-arrival of the incoming signals. It was found that state estimation using TDOA measurements from both UWB sensors has a reduced positioning error compared to the algorithm using TDOA measurements from one UWB sensor, without significantly affecting yaw estimation accuracy. Furthermore, the tag-TDOA measurement did not improve the estimation accuracy at the implemented baseline of 0.22 metres as the<br/>measurement error was too large compared to the baseline.
Alvarez, Marina Gonzalez
Evolved Neuromorphic Altitude Controller for an Autonomous Blimp Masters Thesis
TU Delft Aerospace Engineering, 2021, (Dupeyroux, J.J.G. (mentor); de Croon, G.C.H.E. (graduation committee); Corradi, Federico (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:0cf0f29d-7bdd-4050-817b-2486ed6461d9,
title = {Evolved Neuromorphic Altitude Controller for an Autonomous Blimp},
author = {Marina Gonzalez Alvarez},
url = {http://resolver.tudelft.nl/uuid:0cf0f29d-7bdd-4050-817b-2486ed6461d9},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Micro robotic airships offer significant advantages in terms of safety, mobility, and extended flight times. However, their highly restrictive weight constraints pose a major challenge regarding the available computational power to perform the required control tasks. Thus, spiking neural networks (SNNs) are a promising research direction. By mimicking the biological process for transferring information between neurons using spikes or impulses, they allow for low power consumption and asynchronous event-driven processing. In this work, we propose an evolved altitude controller based on a SNN for an airship which relies solely on the sensory feedback provided by an airborne radar sensor. Starting from the design of a a lightweight, low-cost, open-source airship, we also present a low-control-effort SNN architecture, an evolutionary framework for training the network in a simulated environment, and a control scheme for ameliorating the performance of the system in real-world scenarios. The system's performance is evaluated through real-world experiments, demonstrating the advantages of our approach by comparing it with an artificial neural network (ANN) and a linear controller (PID). The results show an accurate tracking of the altitude command while ensuring efficient management of the control effort. The main contributions of this work are presented in the scientific paper, corresponding to Part I of the document. Besides the research on altitude control based on SNNs and their comparison with an ANN and a PID, this thesis includes an in-depth review of the relevant literate on the main topics covered, in Part II. Finally, a detailed explanation of the methodologies used, the conclusions and recommendations for future work are proposed in Part III.},
note = {Dupeyroux, J.J.G. (mentor); de Croon, G.C.H.E. (graduation committee); Corradi, Federico (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Micro robotic airships offer significant advantages in terms of safety, mobility, and extended flight times. However, their highly restrictive weight constraints pose a major challenge regarding the available computational power to perform the required control tasks. Thus, spiking neural networks (SNNs) are a promising research direction. By mimicking the biological process for transferring information between neurons using spikes or impulses, they allow for low power consumption and asynchronous event-driven processing. In this work, we propose an evolved altitude controller based on a SNN for an airship which relies solely on the sensory feedback provided by an airborne radar sensor. Starting from the design of a a lightweight, low-cost, open-source airship, we also present a low-control-effort SNN architecture, an evolutionary framework for training the network in a simulated environment, and a control scheme for ameliorating the performance of the system in real-world scenarios. The system's performance is evaluated through real-world experiments, demonstrating the advantages of our approach by comparing it with an artificial neural network (ANN) and a linear controller (PID). The results show an accurate tracking of the altitude command while ensuring efficient management of the control effort. The main contributions of this work are presented in the scientific paper, corresponding to Part I of the document. Besides the research on altitude control based on SNNs and their comparison with an ANN and a PID, this thesis includes an in-depth review of the relevant literate on the main topics covered, in Part II. Finally, a detailed explanation of the methodologies used, the conclusions and recommendations for future work are proposed in Part III.
Wang, Sunyi
Thermistor-based airflow sensing on a flapping wing micro air vehicle Masters Thesis
TU Delft Aerospace Engineering, 2021, (van Oudheusden, B.W. (mentor); de Croon, G.C.H.E. (graduation committee); Olejnik, D.A. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:0f908624-ddf3-4329-817e-3170d2b6b656,
title = {Thermistor-based airflow sensing on a flapping wing micro air vehicle},
author = {Sunyi Wang},
url = {http://resolver.tudelft.nl/uuid:0f908624-ddf3-4329-817e-3170d2b6b656},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Flow sensing exists widely in nature to help animals perform certain tasks. It has also been widely adopted in engineering applications with different types of sensing instrumentation. In particular, in the field of aerospace engineering, airflow sensing is crucial to vehicle state evaluation and flight control. This project surveys the key mechanisms from biological features in nature that enable flow sensing and expands towards the application motivation to identify a suitable airflow sensor that can be equipped to a flapping wing micro air vehicle (FWMAV) for onboard airflow sensing. <br/><br/>The selection of sensors is first narrowed down to three major types of airflow sensors from the state of art that have the most potential to be integrated onboard a flapping wing MAV, considering the sensor performance need, size, weight and power (SWaP) restrictions. Two thermal-based commercially available low-cost airflow sensors RevP and RevC from Modern Device have been selected after the trade-off analysis. <br/><br/>A full workflow of calibrating and evaluating the two airflow sensors' directional sensitivity has been carried out through two wind tunnel campaigns. Their performance under grid-generated turbulence is compared with a constant temperature hot-wire anemometer. This series of tests leads to the conclusion that the RevP airflow sensor has better performance and is therefore chosen to be placed onboard a flapping wing MAV Delfly Nimble. <br/><br/>Both mounted tests and tethered hovering tests with the Delfly Nimble are performed to further examine the airflow sensor RevP's measurement performance under different influence factors such as MAV throttle levels, MAV body pitch angles and freestream speeds. In the end, it is concluded that as a proof of concept, the RevP sensor is capable of performing effective measurements for low flow speeds less than 4 m/s, within the pitching angle range of -30 to 30 degrees. Although this is the first achieved tethered hover flight with onboard airflow sensing for a flapping wing MAV, its limited payload and onboard power supply demands an even smaller and less power consuming design of airflow sensors to enable further applications such as autonomous reactive control under wind disturbances.},
note = {van Oudheusden, B.W. (mentor); de Croon, G.C.H.E. (graduation committee); Olejnik, D.A. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Flow sensing exists widely in nature to help animals perform certain tasks. It has also been widely adopted in engineering applications with different types of sensing instrumentation. In particular, in the field of aerospace engineering, airflow sensing is crucial to vehicle state evaluation and flight control. This project surveys the key mechanisms from biological features in nature that enable flow sensing and expands towards the application motivation to identify a suitable airflow sensor that can be equipped to a flapping wing micro air vehicle (FWMAV) for onboard airflow sensing. <br/><br/>The selection of sensors is first narrowed down to three major types of airflow sensors from the state of art that have the most potential to be integrated onboard a flapping wing MAV, considering the sensor performance need, size, weight and power (SWaP) restrictions. Two thermal-based commercially available low-cost airflow sensors RevP and RevC from Modern Device have been selected after the trade-off analysis. <br/><br/>A full workflow of calibrating and evaluating the two airflow sensors' directional sensitivity has been carried out through two wind tunnel campaigns. Their performance under grid-generated turbulence is compared with a constant temperature hot-wire anemometer. This series of tests leads to the conclusion that the RevP airflow sensor has better performance and is therefore chosen to be placed onboard a flapping wing MAV Delfly Nimble. <br/><br/>Both mounted tests and tethered hovering tests with the Delfly Nimble are performed to further examine the airflow sensor RevP's measurement performance under different influence factors such as MAV throttle levels, MAV body pitch angles and freestream speeds. In the end, it is concluded that as a proof of concept, the RevP sensor is capable of performing effective measurements for low flow speeds less than 4 m/s, within the pitching angle range of -30 to 30 degrees. Although this is the first achieved tethered hover flight with onboard airflow sensing for a flapping wing MAV, its limited payload and onboard power supply demands an even smaller and less power consuming design of airflow sensors to enable further applications such as autonomous reactive control under wind disturbances.
Ge, Zhouxin
TU Delft Aerospace Engineering, 2021, (van Kampen, E. (mentor); de Croon, G.C.H.E. (graduation committee); Mitici, M.A. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:d3baec43-71d4-4f7f-ae27-2fdfdae7fea3,
title = {End-to-End Hierarchical Reinforcement Learning for Adaptive Flight Control: A method for model-independent control through Proximal Policy Optimization with learned Options},
author = {Zhouxin Ge},
url = {http://resolver.tudelft.nl/uuid:d3baec43-71d4-4f7f-ae27-2fdfdae7fea3},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Aircraft with disruptive designs have no high-fidelity and accurate flight models. At the same time, developing models for stochastic phenomena for traditional aircraft configurations are costly, and classical control methods cannot operate beyond the predefined operation points or adapt to unexpected changes to the aircraft. The Proximal Policy Option Critic (PPOC) is an end-to-end hierarchical reinforcement learning method that alleviates the need for a high-fidelity flight model and allows for adaptive flight control. This research contributes to the development and analysis of online adaptive flight control by comparing PPOC against a non-hierarchical method called Proximal Policy Optimization (PPO) and PPOC with a single Option (PPOC-1). The methods are tested on an extendable mass-spring-damper system and aircraft model. Subsequently, the agents are evaluated by their sample efficiency, reference tracking capability and adaptivity. The results show, unexpectedly, that PPO and PPOC-1 are more sample efficient than PPOC. Furthermore, both PPOC agents are able to successfully track the height profile, though the agents learn a policy that results in noisy actuator inputs. Finally, PPOC with multiple learned Options has the best adaptivity, as it is able to adapt to structural failure of the horizontal tailplane, sign change of pitch damping, and generalize to different aircraft.},
note = {van Kampen, E. (mentor); de Croon, G.C.H.E. (graduation committee); Mitici, M.A. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Aircraft with disruptive designs have no high-fidelity and accurate flight models. At the same time, developing models for stochastic phenomena for traditional aircraft configurations are costly, and classical control methods cannot operate beyond the predefined operation points or adapt to unexpected changes to the aircraft. The Proximal Policy Option Critic (PPOC) is an end-to-end hierarchical reinforcement learning method that alleviates the need for a high-fidelity flight model and allows for adaptive flight control. This research contributes to the development and analysis of online adaptive flight control by comparing PPOC against a non-hierarchical method called Proximal Policy Optimization (PPO) and PPOC with a single Option (PPOC-1). The methods are tested on an extendable mass-spring-damper system and aircraft model. Subsequently, the agents are evaluated by their sample efficiency, reference tracking capability and adaptivity. The results show, unexpectedly, that PPO and PPOC-1 are more sample efficient than PPOC. Furthermore, both PPOC agents are able to successfully track the height profile, though the agents learn a policy that results in noisy actuator inputs. Finally, PPOC with multiple learned Options has the best adaptivity, as it is able to adapt to structural failure of the horizontal tailplane, sign change of pitch damping, and generalize to different aircraft.
Archundia, Guillermo Gonzalez
Position controller for a flapping-wing drone using ultra wide band Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:b18cead1-951b-4a21-a4bb-0ba36f1768ee,
title = {Position controller for a flapping-wing drone using ultra wide band},
author = {Guillermo Gonzalez Archundia},
url = {http://resolver.tudelft.nl/uuid:b18cead1-951b-4a21-a4bb-0ba36f1768ee},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {The continuous improvement and miniaturisation of elements in drones have been essential for making flapping-wing drones a reality. This thesis presents an integral approach for accurate indoor position control and estimation on flapping-wing drones. The approach considers three main aspects to enhance transient response of the drone. The first one is an experimental velocity/attitude flapping-wing model for drag compensation, obtained through system identification techniques. The second one is a voltage-dependent variable thrust model for enhancing height control. Thirdly, a characterisation of ground effects to determine the height for stable hovering. For the state estimation, an extended Kalman filter fuses UWB position measurements with IMU data. Due to the well-known multi-path effects of UWB, the Kalman filter includes an adaptive noise parameter based on height. The novel control strategy was validated with real flight tests, where position control improved by a factor of 1.5, reaching a mean absolute error of 10cm in positions in x and y, and 4.9cm for position in z.},
note = {de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
The continuous improvement and miniaturisation of elements in drones have been essential for making flapping-wing drones a reality. This thesis presents an integral approach for accurate indoor position control and estimation on flapping-wing drones. The approach considers three main aspects to enhance transient response of the drone. The first one is an experimental velocity/attitude flapping-wing model for drag compensation, obtained through system identification techniques. The second one is a voltage-dependent variable thrust model for enhancing height control. Thirdly, a characterisation of ground effects to determine the height for stable hovering. For the state estimation, an extended Kalman filter fuses UWB position measurements with IMU data. Due to the well-known multi-path effects of UWB, the Kalman filter includes an adaptive noise parameter based on height. The novel control strategy was validated with real flight tests, where position control improved by a factor of 1.5, reaching a mean absolute error of 10cm in positions in x and y, and 4.9cm for position in z.
Vroon, Erik
Motion-based MAV Detection in GPS-denied Environments Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); Rojer, Jim (mentor); Guo, J. (graduation committee); de Visser, C.C. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:9ddca1f8-3956-42b7-a683-b418fcd89df2,
title = {Motion-based MAV Detection in GPS-denied Environments},
author = {Erik Vroon},
url = {http://resolver.tudelft.nl/uuid:9ddca1f8-3956-42b7-a683-b418fcd89df2},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Drones need to be able to detect and localize each other if they are to collaborate in multi-robot teams or swarms. Typically, computer vision methods based on visual appearance are investigated to this end. In contrast, in this work, a method based on dense optical flow (OF) is developed that detects dynamic objects. This is achieved by comparing the flow vectors of dense OF with the direction to the Focus of Expansion (FoE) in the image plane. A simulation in AirSim is developed to validate this approach and to create datasets for motion-based object detection of MAVs. This simulation includes ground-truth FoE, depth, OF and IMU data. The results show that this method performs well if the OF vector's magnitude is large enough and its angle is sufficiently different from those of static world points. We expect that the presented method will serve as a useful baseline for deep learning methods that use dense optical flow as input.},
note = {de Croon, G.C.H.E. (mentor); Rojer, Jim (mentor); Guo, J. (graduation committee); de Visser, C.C. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Drones need to be able to detect and localize each other if they are to collaborate in multi-robot teams or swarms. Typically, computer vision methods based on visual appearance are investigated to this end. In contrast, in this work, a method based on dense optical flow (OF) is developed that detects dynamic objects. This is achieved by comparing the flow vectors of dense OF with the direction to the Focus of Expansion (FoE) in the image plane. A simulation in AirSim is developed to validate this approach and to create datasets for motion-based object detection of MAVs. This simulation includes ground-truth FoE, depth, OF and IMU data. The results show that this method performs well if the OF vector's magnitude is large enough and its angle is sufficiently different from those of static world points. We expect that the presented method will serve as a useful baseline for deep learning methods that use dense optical flow as input.
Keltjens, Benjamin
Self-Supervised Monocular Depth Estimation of Untextured Indoor Rotated Scenes Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); van Dijk, Tom (mentor); van Gemert, J.C. (graduation committee); Smeur, E.J.J. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:9b68db7c-ac32-422e-8749-a8e0bc1fc4ca,
title = {Self-Supervised Monocular Depth Estimation of Untextured Indoor Rotated Scenes},
author = {Benjamin Keltjens},
url = {http://resolver.tudelft.nl/uuid:9b68db7c-ac32-422e-8749-a8e0bc1fc4ca},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Self-supervised deep learning methods have leveraged stereo images for training monocular depth estimation. Although these methods show strong results on outdoor datasets such as KITTI, they do not match performance of supervised methods on indoor environments with camera rotation. Indoor, rotated scenes are common for less constrained applications and pose problems for two reasons: abundance of low texture regions and increased complexity of depth cues for images under rotation. In an effort to extend self-supervised learning to more generalised environments we propose two additions. First, we propose a novel Filled Disparity Loss term that corrects for ambiguity of image reconstruction error loss in textureless regions. Specifically, we interpolate disparity in untextured regions, using the estimated disparity from surrounding textured areas, and use L1 loss to correct the original estimation. Our experiments show that depth estimation is substantially improved on low-texture scenes, without any loss on textured scenes, when compared to Monodepth by Godard et al. Secondly, we show that training with an application's representative rotations, in both pitch and roll, is sufficient to significantly improve performance over the entire range of expected rotation. We demonstrate that depth estimation is successfully generalised as performance is not lost when evaluated on test sets with no camera rotation. Together these developments enable a broader use of self-supervised learning of monocular depth estimation for complex environments.},
note = {de Croon, G.C.H.E. (mentor); van Dijk, Tom (mentor); van Gemert, J.C. (graduation committee); Smeur, E.J.J. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Self-supervised deep learning methods have leveraged stereo images for training monocular depth estimation. Although these methods show strong results on outdoor datasets such as KITTI, they do not match performance of supervised methods on indoor environments with camera rotation. Indoor, rotated scenes are common for less constrained applications and pose problems for two reasons: abundance of low texture regions and increased complexity of depth cues for images under rotation. In an effort to extend self-supervised learning to more generalised environments we propose two additions. First, we propose a novel Filled Disparity Loss term that corrects for ambiguity of image reconstruction error loss in textureless regions. Specifically, we interpolate disparity in untextured regions, using the estimated disparity from surrounding textured areas, and use L1 loss to correct the original estimation. Our experiments show that depth estimation is substantially improved on low-texture scenes, without any loss on textured scenes, when compared to Monodepth by Godard et al. Secondly, we show that training with an application's representative rotations, in both pitch and roll, is sufficient to significantly improve performance over the entire range of expected rotation. We demonstrate that depth estimation is successfully generalised as performance is not lost when evaluated on test sets with no camera rotation. Together these developments enable a broader use of self-supervised learning of monocular depth estimation for complex environments.
Dvorsky, Nicholas
Feasibility of using electric drone main rotors for electricity generation vs. solar panels for indefinite flight Masters Thesis
TU Delft Electrical Engineering, Mathematics and Computer Science, 2021, (Zaaijer, M B (mentor); de Croon, G.C.H.E. (graduation committee); Schmehl, R. (graduation committee); Remes, B.D.W. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:5aa0ed78-c775-4a1b-a0ab-5145f85e5e9d,
title = {Feasibility of using electric drone main rotors for electricity generation vs. solar panels for indefinite flight},
author = {Nicholas Dvorsky},
url = {http://resolver.tudelft.nl/uuid:5aa0ed78-c775-4a1b-a0ab-5145f85e5e9d},
year = {2021},
date = {2021-01-01},
school = {TU Delft Electrical Engineering, Mathematics and Computer Science},
abstract = {An idea was proposed to allow an autonomous drone to have indefinite flight times over the ocean by applying renewable energy technologies and theory to generate electricity in flight. This is considered less as a way to save energy, but to permit the use of such a drone from a ship not capable of safely retrieving it. One novel component of this idea is to use the wind updraft created by the motion of a ship or natural air currents as the wind source for an on-board turbine generator. The second component is to use the existing drive system as the on-board turbine in a 'hybrid rotor' design to reduce the need for extra parts and complexity. This report analyzes the potential for such a system compared to a more intuitive airborne solar system, and to the combination of both concepts. While indefinite flight time is paramount, the goal is to maximize the "mission" time to charge/idle time ratio. The process for determining fitness is a simulation of the aircraft flying on its mission and charging when needed (and if possible) for a full year for varying designs of aircraft and rotor. The results of all the tests show that the main idea is infeasible because not enough energy can be generated from the inefficient propeller and the updrafts are insufficient and inconsistent. The alternatives of solar and combined power systems function better but are still subject to high failure rates. The most promising system is to use a separate turbine and propeller and also include solar panels to achieve the most effectiveness both when in powered flight and while charging. This constitutes a compromise on the 'hybrid rotor' part of the idea. The conclusion of this report is that further improvements to the design and control of the most successful configuration are possible could result in a fully functional system.},
note = {Zaaijer, M B (mentor); de Croon, G.C.H.E. (graduation committee); Schmehl, R. (graduation committee); Remes, B.D.W. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
An idea was proposed to allow an autonomous drone to have indefinite flight times over the ocean by applying renewable energy technologies and theory to generate electricity in flight. This is considered less as a way to save energy, but to permit the use of such a drone from a ship not capable of safely retrieving it. One novel component of this idea is to use the wind updraft created by the motion of a ship or natural air currents as the wind source for an on-board turbine generator. The second component is to use the existing drive system as the on-board turbine in a 'hybrid rotor' design to reduce the need for extra parts and complexity. This report analyzes the potential for such a system compared to a more intuitive airborne solar system, and to the combination of both concepts. While indefinite flight time is paramount, the goal is to maximize the "mission" time to charge/idle time ratio. The process for determining fitness is a simulation of the aircraft flying on its mission and charging when needed (and if possible) for a full year for varying designs of aircraft and rotor. The results of all the tests show that the main idea is infeasible because not enough energy can be generated from the inefficient propeller and the updrafts are insufficient and inconsistent. The alternatives of solar and combined power systems function better but are still subject to high failure rates. The most promising system is to use a separate turbine and propeller and also include solar panels to achieve the most effectiveness both when in powered flight and while charging. This constitutes a compromise on the 'hybrid rotor' part of the idea. The conclusion of this report is that further improvements to the design and control of the most successful configuration are possible could result in a fully functional system.
Roulaux, Bas
Attitude Control of Flapping-Wing Air Vehicles Masters Thesis
TU Delft Mechanical, Maritime and Materials Engineering; TU Delft Precision and Microsystems Engineering, 2021, (Goosen, J.F.L. (mentor); Remes, B.D.W. (graduation committee); van der Wijk, V. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:f46c66f3-60e5-4285-9cf3-978826387526,
title = {Attitude Control of Flapping-Wing Air Vehicles},
author = {Bas Roulaux},
url = {http://resolver.tudelft.nl/uuid:f46c66f3-60e5-4285-9cf3-978826387526},
year = {2021},
date = {2021-01-01},
school = {TU Delft Mechanical, Maritime and Materials Engineering; TU Delft Precision and Microsystems Engineering},
abstract = {Flapping-Wing Air Vehicles (FWAV) are autonomously flying vehicles that use their flapping wings to simultaneously stay aloft and enable controllable flight. FWAVs that are capable of controllable flight are reported in literature, though a theoretical background of the aerodynamic performance of different attitude control mechanisms is absent in literature and the robustness of attitude control mechanisms with respect to body motions is oftentimes omitted. The aim of this thesis is to develop a theoretical framework for the aerodynamic response of flapping wings that includes variation of attitude control parameters and motion of the vehicle body. This framework can be used to assist in research into new attitude control mechanisms for FWAVs that are not yet capable of attitude control, such as the compliant Atalanta FWAV. Analytical aerodynamic and kinematic descriptions are combined to analyze the aerodynamic performance of two suggested attitude control mechanisms: stroke amplitude variations and control of the angle of attack by means of pitching stiffness variations. It is shown in this research that both mechanisms have a significant influence on the lift production of a flapping wing, though this influence changes significantly when body motions are introduced. It is found that variations of the stroke amplitude provide the most predictable variations in lift for all cases of body motion that were considered, provided that the wing’s pitching hinge stiffness is high enough to ensure stable flapping kinematics under the influence of body motion.},
note = {Goosen, J.F.L. (mentor); Remes, B.D.W. (graduation committee); van der Wijk, V. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Flapping-Wing Air Vehicles (FWAV) are autonomously flying vehicles that use their flapping wings to simultaneously stay aloft and enable controllable flight. FWAVs that are capable of controllable flight are reported in literature, though a theoretical background of the aerodynamic performance of different attitude control mechanisms is absent in literature and the robustness of attitude control mechanisms with respect to body motions is oftentimes omitted. The aim of this thesis is to develop a theoretical framework for the aerodynamic response of flapping wings that includes variation of attitude control parameters and motion of the vehicle body. This framework can be used to assist in research into new attitude control mechanisms for FWAVs that are not yet capable of attitude control, such as the compliant Atalanta FWAV. Analytical aerodynamic and kinematic descriptions are combined to analyze the aerodynamic performance of two suggested attitude control mechanisms: stroke amplitude variations and control of the angle of attack by means of pitching stiffness variations. It is shown in this research that both mechanisms have a significant influence on the lift production of a flapping wing, though this influence changes significantly when body motions are introduced. It is found that variations of the stroke amplitude provide the most predictable variations in lift for all cases of body motion that were considered, provided that the wing’s pitching hinge stiffness is high enough to ensure stable flapping kinematics under the influence of body motion.
Wessendorp, Nikhil
Obstacle Avoidance onboard MAVs using a FMCW RADAR Masters Thesis
TU Delft Aerospace Engineering, 2021, (Dupeyroux, J.J.G. (mentor); de Croon, G.C.H.E. (mentor); Fioranelli, F. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:019566ca-34de-4ddd-87ac-86364ef2759b,
title = {Obstacle Avoidance onboard MAVs using a FMCW RADAR},
author = {Nikhil Wessendorp},
url = {http://resolver.tudelft.nl/uuid:019566ca-34de-4ddd-87ac-86364ef2759b},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Micro air vehicles (MAVs) are increasingly being considered for aerial tasks such as delivery of goods and surveillance due to their lightweight, compact design and manoeuvrability. To safely and reliably carry out these tasks and navigate to its objective, especially in complex and cluttered environments, the MAV is also required to sense and avoid (S&A) obstacles. Due to the MAVs limitations in weight, power and processing power, vision systems usually prove ideal for sensing the environment, being a cheap, lightweight, power efficient and a rich source of information. They do however require adequate computational resources and most importantly, good visibility. When the environment does not host these conditions, for instance when flying though dust, smoke or fog, other sensors need to be utilised that can provide more robust sensing to ensure safe and reliable operation. Radar sensors are mostly unaffected by atmospheric conditions and have been used extensively in the aerospace industry for this purpose. These sensors were traditionally heavy and power hungry, only applicable on ground or in large craft. However other radar sensors have since come about that are more suited for use in small MAVs. Specifically, lightweight, power efficient and compact frequency modulated continuous wave (FMCW) radars have increasingly been used in advanced driver assistance systems as auxiliary sensors, however there has been little work to integrate them on MAVs. This sensor provides the range, horizontal bearing and radial velocity (Doppler shift) of any objects in the field of view, which can then be used for multi-target tracking (MTT) [38]. The major disadvantage of the sensor is the limited field of view (approximately 80 degrees horizontal) and noisy nature of the sensor, especially in cluttered environments. The challenge is to explore filtering, tracking and avoidance algorithm pipelines to extract meaningful information from the raw data and investigate the sensor’s effectiveness with respect to obstacle avoidance on MAVs. This will include algorithms such as data association, estimation and avoidance, as well as an investigation of neural networks to aid in processing the raw data and provide some filtering. This will be accomplished by integrating the sensor on a MAV and testing and tuning the algorithms both in real life (in the cyberzoo flying arena of the aerospace faculty), and using data gathered as part of an obstacle detection and avoidance dataset that was generated during this project. This will hopefully allow MAVs to operate safer, either using a standalone radar or integrated with other sensors.},
note = {Dupeyroux, J.J.G. (mentor); de Croon, G.C.H.E. (mentor); Fioranelli, F. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Micro air vehicles (MAVs) are increasingly being considered for aerial tasks such as delivery of goods and surveillance due to their lightweight, compact design and manoeuvrability. To safely and reliably carry out these tasks and navigate to its objective, especially in complex and cluttered environments, the MAV is also required to sense and avoid (S&A) obstacles. Due to the MAVs limitations in weight, power and processing power, vision systems usually prove ideal for sensing the environment, being a cheap, lightweight, power efficient and a rich source of information. They do however require adequate computational resources and most importantly, good visibility. When the environment does not host these conditions, for instance when flying though dust, smoke or fog, other sensors need to be utilised that can provide more robust sensing to ensure safe and reliable operation. Radar sensors are mostly unaffected by atmospheric conditions and have been used extensively in the aerospace industry for this purpose. These sensors were traditionally heavy and power hungry, only applicable on ground or in large craft. However other radar sensors have since come about that are more suited for use in small MAVs. Specifically, lightweight, power efficient and compact frequency modulated continuous wave (FMCW) radars have increasingly been used in advanced driver assistance systems as auxiliary sensors, however there has been little work to integrate them on MAVs. This sensor provides the range, horizontal bearing and radial velocity (Doppler shift) of any objects in the field of view, which can then be used for multi-target tracking (MTT) [38]. The major disadvantage of the sensor is the limited field of view (approximately 80 degrees horizontal) and noisy nature of the sensor, especially in cluttered environments. The challenge is to explore filtering, tracking and avoidance algorithm pipelines to extract meaningful information from the raw data and investigate the sensor’s effectiveness with respect to obstacle avoidance on MAVs. This will include algorithms such as data association, estimation and avoidance, as well as an investigation of neural networks to aid in processing the raw data and provide some filtering. This will be accomplished by integrating the sensor on a MAV and testing and tuning the algorithms both in real life (in the cyberzoo flying arena of the aerospace faculty), and using data gathered as part of an obstacle detection and avoidance dataset that was generated during this project. This will hopefully allow MAVs to operate safer, either using a standalone radar or integrated with other sensors.
Booster, Quincy
Urban MAV: A visual odometry dataset Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:a4497602-a257-4561-8e4c-5baa36e8cd6f,
title = {Urban MAV: A visual odometry dataset},
author = {Quincy Booster},
url = {http://resolver.tudelft.nl/uuid:a4497602-a257-4561-8e4c-5baa36e8cd6f},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {The most common used classification of drones is Micro air vehicles (MAV), with quadrotors being the most conventional MAV [1]. Its relatively small size and rotary wings provide high maneuverability, including vertical takeoff and hovering, making them useful in confined and hard to reach spaces. Another benefit of the MAV in comparison to other drone classifications is its relatively smaller production costs [1]. Because of these features they play an increasing role within our society by aiding in human tasks, by applying them in for example site inspection, agriculture and rescue missions [2–4]. However, its agility comes at a cost which takes the form of high power consumption [5]. Making improving MAV designs a challenge.},
note = {de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
The most common used classification of drones is Micro air vehicles (MAV), with quadrotors being the most conventional MAV [1]. Its relatively small size and rotary wings provide high maneuverability, including vertical takeoff and hovering, making them useful in confined and hard to reach spaces. Another benefit of the MAV in comparison to other drone classifications is its relatively smaller production costs [1]. Because of these features they play an increasing role within our society by aiding in human tasks, by applying them in for example site inspection, agriculture and rescue missions [2–4]. However, its agility comes at a cost which takes the form of high power consumption [5]. Making improving MAV designs a challenge.
Chotalal, Rohan
High-Dimensional Optimal State-Feedback Mapping using Deep Neural Networks for Agile Quadrotor Flight Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Wagter, C. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:8f90894b-933a-4649-90a0-1bbc1de6c0d6,
title = {High-Dimensional Optimal State-Feedback Mapping using Deep Neural Networks for Agile Quadrotor Flight},
author = {Rohan Chotalal},
url = {http://resolver.tudelft.nl/uuid:8f90894b-933a-4649-90a0-1bbc1de6c0d6},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {For most robotics applications, optimal control remains a promising solution for solving complex control tasks. One example is the time-optimal flight of Micro Air Vehicles (MAVs), where strict computational requirements fail to resolve such algorithms onboard. Recent work on the use of deep neural networks for guidance and control (G&CNets) has shown that these biologically inspired models approximate well the optimal control solution while requiring a fraction of the computational cost. Although previous attempts resulted in successful flight tests, training occurred on large-scale datasets based on a 3-DoF model. Since model refinement leads to higher generation time, in this work, we show that G&CNets trained on small-sized datasets can mimic the optimal control solution of a full 6-DoF quadrotor model. The cost function used in the generation process penalizes the altitude error and mixes both time and power-optimal objectives weighted by a varying homotopy parameter. Trained networks output the vertical thrust command and body rates based on the vehicle's position, velocity, and attitude. The proposed controller transfers well onboard for different flight scenarios: (i) longitudinal, lateral and diagonal flight; (ii) hovering with and without the effect of disturbances and (iii) waypoint tracking experiment. Through a Monte-Carlo test campaign, it is demonstrated that G&CNets trained on small datasets provide similar results to those with 100 times more samples. To the best of our knowledge, this work is the first implementation of a high-dimensional G&CNet in the control loop of a real MAV.},
note = {de Wagter, C. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
For most robotics applications, optimal control remains a promising solution for solving complex control tasks. One example is the time-optimal flight of Micro Air Vehicles (MAVs), where strict computational requirements fail to resolve such algorithms onboard. Recent work on the use of deep neural networks for guidance and control (G&CNets) has shown that these biologically inspired models approximate well the optimal control solution while requiring a fraction of the computational cost. Although previous attempts resulted in successful flight tests, training occurred on large-scale datasets based on a 3-DoF model. Since model refinement leads to higher generation time, in this work, we show that G&CNets trained on small-sized datasets can mimic the optimal control solution of a full 6-DoF quadrotor model. The cost function used in the generation process penalizes the altitude error and mixes both time and power-optimal objectives weighted by a varying homotopy parameter. Trained networks output the vertical thrust command and body rates based on the vehicle's position, velocity, and attitude. The proposed controller transfers well onboard for different flight scenarios: (i) longitudinal, lateral and diagonal flight; (ii) hovering with and without the effect of disturbances and (iii) waypoint tracking experiment. Through a Monte-Carlo test campaign, it is demonstrated that G&CNets trained on small datasets provide similar results to those with 100 times more samples. To the best of our knowledge, this work is the first implementation of a high-dimensional G&CNet in the control loop of a real MAV.
Dinaux, Raoul
Obstacle Detection and Avoidance onboard an MAV using a Monocular Event-based Camera Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); Dupeyroux, J.J.G. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:85b358c9-8018-4e0c-867e-f35fe26716cb,
title = {Obstacle Detection and Avoidance onboard an MAV using a Monocular Event-based Camera},
author = {Raoul Dinaux},
url = {http://resolver.tudelft.nl/uuid:85b358c9-8018-4e0c-867e-f35fe26716cb},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Micro Air Vehicles (MAVs) are able to support humans in dangerous operations, such as search and rescue operations at night on unknown terrain. These scenes require a great amount of autonomy from the MAV, as they are often radio and GPS-denied. As MAVs have limited computational resources and energy storage, onboard navigation tasks have to be performed efficient and fast. To address this challenge, this research proposes an approach to visual obstacle detection and avoidance onboard an MAV. The algorithmic approach is based on event-based optic flow, using a monocular event-based camera. This camera captures the apparent motion in the scene, has microsecond latency and very low power consumption, therefore a good fit for onboard navigation tasks. Firstly, a literature study is performed to provide theoretical concepts and foundation for the obstacle avoidance approach. A processing pipeline is designed, based on the use of event-based normal optic flow. This pipeline consists of three sections: course estimation, obstacle detection and obstacle avoidance. A novel course estimation method 'FAITH' is proposed which uses optic flow half-planes along with a fast RANSAC scheme. The object detection method is based on DBSCAN clustering of optic flow vectors, using the time-to-contact and vector location as clustering variables. The performance of these methods is experimentally demonstrated by three experiments: in a simulated environment, offline on real sensor data and online onboard an MAV. As currently no event-based obstacle avoidance datasets are publicly available, a dataset is recorded as supplement to this and future research. Approximately 1350 runs of event-based camera, RADAR, IMU and OptiTrack data are recorded, manually avoiding either a single or two poles using an MAV in the flying arena of the TU Delft. This dataset is used in this research to determine the performance of the course estimation method using real sensor data. The course estimation method is shown to have state-of-the-art accuracy and beyond state-of-the-art computation time on both simulated data and the recorded dataset. The final experiment shows the obstacle detection and avoidance approach integrated onboard an MAV in a real-time obstacle avoidance task. The approach is shown to have a success rate of 80% in a frontal obstacle avoidance task on a low-textured 50-cm wide pole. The contribution of this research is an obstacle detection and avoidance approach using a monocular event-based camera onboard an MAV, along with the novel course estimation algorithm 'FAITH'.},
note = {de Croon, G.C.H.E. (mentor); Dupeyroux, J.J.G. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Micro Air Vehicles (MAVs) are able to support humans in dangerous operations, such as search and rescue operations at night on unknown terrain. These scenes require a great amount of autonomy from the MAV, as they are often radio and GPS-denied. As MAVs have limited computational resources and energy storage, onboard navigation tasks have to be performed efficient and fast. To address this challenge, this research proposes an approach to visual obstacle detection and avoidance onboard an MAV. The algorithmic approach is based on event-based optic flow, using a monocular event-based camera. This camera captures the apparent motion in the scene, has microsecond latency and very low power consumption, therefore a good fit for onboard navigation tasks. Firstly, a literature study is performed to provide theoretical concepts and foundation for the obstacle avoidance approach. A processing pipeline is designed, based on the use of event-based normal optic flow. This pipeline consists of three sections: course estimation, obstacle detection and obstacle avoidance. A novel course estimation method 'FAITH' is proposed which uses optic flow half-planes along with a fast RANSAC scheme. The object detection method is based on DBSCAN clustering of optic flow vectors, using the time-to-contact and vector location as clustering variables. The performance of these methods is experimentally demonstrated by three experiments: in a simulated environment, offline on real sensor data and online onboard an MAV. As currently no event-based obstacle avoidance datasets are publicly available, a dataset is recorded as supplement to this and future research. Approximately 1350 runs of event-based camera, RADAR, IMU and OptiTrack data are recorded, manually avoiding either a single or two poles using an MAV in the flying arena of the TU Delft. This dataset is used in this research to determine the performance of the course estimation method using real sensor data. The course estimation method is shown to have state-of-the-art accuracy and beyond state-of-the-art computation time on both simulated data and the recorded dataset. The final experiment shows the obstacle detection and avoidance approach integrated onboard an MAV in a real-time obstacle avoidance task. The approach is shown to have a success rate of 80% in a frontal obstacle avoidance task on a low-textured 50-cm wide pole. The contribution of this research is an obstacle detection and avoidance approach using a monocular event-based camera onboard an MAV, along with the novel course estimation algorithm 'FAITH'.
Westenberger, Jelle
Time-Optimal Control for Tiny Quadcopters Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); de Wagter, C. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:665779ce-5080-43da-8519-4cd17e2f105d,
title = {Time-Optimal Control for Tiny Quadcopters},
author = {Jelle Westenberger},
url = {http://resolver.tudelft.nl/uuid:665779ce-5080-43da-8519-4cd17e2f105d},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Time-optimal model-predictive control is essential in achieving fast and adaptive quadcopter flight. Due to the limited computational performance of onboard hardware, aggressive flight approaches have relied on off-line trajectory optimization processes or non time-optimal methods. In this work we propose a computational efficient model predictive controller (MPC) that approaches time-optimal flight and runs onboard a consumer quadcopter. The proposed controller is built on the principle that constrained optimal control problems (OCPs) have a so-called 'bang-bang' solution. Our solution plans a bang-bang maneuver in the critical direction while aiming for a 'minimum-effort' approach in non-critical direction. Control parameters are computed by means of a bisection scheme using an analytical path prediction model. The controller has been compared with a classical PID controller and theoretical time-optimal trajectories in simulations. We identify the consequences of the OCP simplifications and propose a method to mitigate one of these effects. Finally, we have implemented the proposed controller onboard a consumer quadcopter and performed indoor flights to compare the controller's performance to a PID controller. Flight experiments have shown that the controller runs at 512hz onboard a Parrot Bebop quadcopter and is capable of fast, saturated flight, outperforming traditional PID controllers in waypoint-to-waypoint flight while requiring only minimal knowledge of the quadcopter's dynamics.},
note = {de Croon, G.C.H.E. (mentor); de Wagter, C. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Time-optimal model-predictive control is essential in achieving fast and adaptive quadcopter flight. Due to the limited computational performance of onboard hardware, aggressive flight approaches have relied on off-line trajectory optimization processes or non time-optimal methods. In this work we propose a computational efficient model predictive controller (MPC) that approaches time-optimal flight and runs onboard a consumer quadcopter. The proposed controller is built on the principle that constrained optimal control problems (OCPs) have a so-called 'bang-bang' solution. Our solution plans a bang-bang maneuver in the critical direction while aiming for a 'minimum-effort' approach in non-critical direction. Control parameters are computed by means of a bisection scheme using an analytical path prediction model. The controller has been compared with a classical PID controller and theoretical time-optimal trajectories in simulations. We identify the consequences of the OCP simplifications and propose a method to mitigate one of these effects. Finally, we have implemented the proposed controller onboard a consumer quadcopter and performed indoor flights to compare the controller's performance to a PID controller. Flight experiments have shown that the controller runs at 512hz onboard a Parrot Bebop quadcopter and is capable of fast, saturated flight, outperforming traditional PID controllers in waypoint-to-waypoint flight while requiring only minimal knowledge of the quadcopter's dynamics.
Woude, Mark
Acoustic-Based Aircraft Detection and Ego-Noise Suppression: for Micro Aerial Vehicles Masters Thesis
TU Delft Aerospace Engineering, 2021, (van Dijk, Tom (mentor); de Croon, G.C.H.E. (mentor); Snellen, M. (graduation committee); de Wagter, C. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:9f478516-7129-4e6f-9b84-d66fad419d51,
title = {Acoustic-Based Aircraft Detection and Ego-Noise Suppression: for Micro Aerial Vehicles},
author = {Mark Woude},
url = {http://resolver.tudelft.nl/uuid:9f478516-7129-4e6f-9b84-d66fad419d51},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Widespread usage of Micro Aerial Vehicles (MAVs) has led to various airspace safety breaches, including near mid-air collisions with other aircraft. To ensure safe integration into general aviation, it is paramount that MAVs are equipped with an autonomous detect and avoid system when flying beyond the visual line-of-sight of the operator. The purpose of this research is to investigate the feasibility of acoustic-based aircraft detection, which has generally been overlooked in favor of optical or radar-based technology. Effective sound-based aircraft detection on-board an MAV requires suppressing the dynamic ego-noise it generates during flight, which would otherwise pollute the recorded environmental sound. This paper proposes using a recurrent neural network to predict the generated noise, given a sequence of MAV flight data, so that it can be effectively removed from noisy recordings. For aircraft detection, a convolutional neural network in combination with Mel spectrogram features is designed to classify noise-free environmental sound as either aircraft or non-aircraft, achieving 97.5% accuracy. To reconstruct the noisy environment of an MAV flight, these noise-free sounds are mixed with ego-noise at mix ratios up to 1.00. When evaluating in these mismatched conditions, accuracy decreases to 95.0% and 47.5% with- and without ego-noise suppression, respectively. Although ego-noise suppression can not prevent a drop in performance, the large difference between the mismatched conditions does demonstrate the benefits of the proposed denoising approach on aircraft detection.},
note = {van Dijk, Tom (mentor); de Croon, G.C.H.E. (mentor); Snellen, M. (graduation committee); de Wagter, C. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Widespread usage of Micro Aerial Vehicles (MAVs) has led to various airspace safety breaches, including near mid-air collisions with other aircraft. To ensure safe integration into general aviation, it is paramount that MAVs are equipped with an autonomous detect and avoid system when flying beyond the visual line-of-sight of the operator. The purpose of this research is to investigate the feasibility of acoustic-based aircraft detection, which has generally been overlooked in favor of optical or radar-based technology. Effective sound-based aircraft detection on-board an MAV requires suppressing the dynamic ego-noise it generates during flight, which would otherwise pollute the recorded environmental sound. This paper proposes using a recurrent neural network to predict the generated noise, given a sequence of MAV flight data, so that it can be effectively removed from noisy recordings. For aircraft detection, a convolutional neural network in combination with Mel spectrogram features is designed to classify noise-free environmental sound as either aircraft or non-aircraft, achieving 97.5% accuracy. To reconstruct the noisy environment of an MAV flight, these noise-free sounds are mixed with ego-noise at mix ratios up to 1.00. When evaluating in these mismatched conditions, accuracy decreases to 95.0% and 47.5% with- and without ego-noise suppression, respectively. Although ego-noise suppression can not prevent a drop in performance, the large difference between the mismatched conditions does demonstrate the benefits of the proposed denoising approach on aircraft detection.
Buysscher, Diego De
Safe Curriculum Learning for Linear Systems With Unknown Dynamics in Primary Flight Control Masters Thesis
TU Delft Aerospace Engineering, 2021, (de Croon, G.C.H.E. (mentor); van Kampen, E. (graduation committee); Mooij, E. (graduation committee); Pollack, T.S.C. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:022cdfe2-737e-44f4-8497-0159a826b9ea,
title = {Safe Curriculum Learning for Linear Systems With Unknown Dynamics in Primary Flight Control},
author = {Diego De Buysscher},
url = {http://resolver.tudelft.nl/uuid:022cdfe2-737e-44f4-8497-0159a826b9ea},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Safe Curriculum Learning constitutes a collection of methods that aim at enabling Rein- forcement Learning (RL) algorithms on complex systems and tasks whilst considering the safety and efficiency aspect of the learning process. On the one hand, curricular reinforce- ment learning approaches divide the task into more gradual complexity stages to promote learning efficiency. On the other, safe learning provides a framework to consider a system’s safety during the learning process. The latter’s contribution is significant on safety-critical systems, such as transport vehicles where stringent (safety) requirements apply. This pa- per proposes a black box safe curriculum learning architecture applicable to systems with unknown dynamics. It only requires knowledge of the state and action spaces’ orders for a given task and system. By adding system identification capabilities to existing safe cur- riculum learning paradigms, the proposed architecture successfully ensures safe learning proceedings of tracking tasks without requiring initial knowledge of internal system dynam- ics. More specifically, a model estimate is generated online to complement safety filters that rely on uncertain models for their safety guarantees. This research explicitly targets linearised systems with decoupled dynamics in the experiments provided in this article as proof of concept. The paradigm is initially verified on a mass-spring-damper system. After that, the architecture is applied to a quadrotor where it is able to successfully track the system’s four degrees of freedom independently, namely attitude angles and altitude. The RL agent is able to safely learn an optimal policy that can track an independent reference on each degree of freedom.},
note = {de Croon, G.C.H.E. (mentor); van Kampen, E. (graduation committee); Mooij, E. (graduation committee); Pollack, T.S.C. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Safe Curriculum Learning constitutes a collection of methods that aim at enabling Rein- forcement Learning (RL) algorithms on complex systems and tasks whilst considering the safety and efficiency aspect of the learning process. On the one hand, curricular reinforce- ment learning approaches divide the task into more gradual complexity stages to promote learning efficiency. On the other, safe learning provides a framework to consider a system’s safety during the learning process. The latter’s contribution is significant on safety-critical systems, such as transport vehicles where stringent (safety) requirements apply. This pa- per proposes a black box safe curriculum learning architecture applicable to systems with unknown dynamics. It only requires knowledge of the state and action spaces’ orders for a given task and system. By adding system identification capabilities to existing safe cur- riculum learning paradigms, the proposed architecture successfully ensures safe learning proceedings of tracking tasks without requiring initial knowledge of internal system dynam- ics. More specifically, a model estimate is generated online to complement safety filters that rely on uncertain models for their safety guarantees. This research explicitly targets linearised systems with decoupled dynamics in the experiments provided in this article as proof of concept. The paradigm is initially verified on a mass-spring-damper system. After that, the architecture is applied to a quadrotor where it is able to successfully track the system’s four degrees of freedom independently, namely attitude angles and altitude. The RL agent is able to safely learn an optimal policy that can track an independent reference on each degree of freedom.
Geursen, Izaak
Fleet Planning Under Demand and Fuel Price Uncertainty Using Actor-Critic Reinforcement Learning Masters Thesis
TU Delft Aerospace Engineering, 2021, (Lopes Dos Santos, Bruno (mentor); Yorke-Smith, Neil (graduation committee); de Croon, Guido (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:1eed4184-5c4a-4e88-b54d-8c9751f79ebf,
title = {Fleet Planning Under Demand and Fuel Price Uncertainty Using Actor-Critic Reinforcement Learning},
author = {Izaak Geursen},
url = {http://resolver.tudelft.nl/uuid:1eed4184-5c4a-4e88-b54d-8c9751f79ebf},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Current state-of-the-art airline planning models are required to decrease models either in size or complexity due to computational limitations, limiting the <br/>operational applicability to problems of representative sizes. Models return suboptimal solutions, especially when confronted with factors of uncertainty. Considering the growing interest in the application of Machine Learning techniques in the Operations Research domain, and the proven success in other fields such as robotics, this research investigates the applicability of these techniques for airline planning. An Advantage Actor-Critic (A2C) Reinforcement Learning agent is applied to the airline fleet planning problem. Because of the increased computational efficiency of using an A2C agent, the problem is increased in size and the highly volatile uncertainty in fuel price is implemented.<br/>Conversion was achieved, and when evaluating the quality of the solutions compared to a deterministic model, the performance was very satisfactory. The A2C agent was able to outperform the deterministic model, with an increasing performance as more complexity was added to the problem. It was found that<br/>the introduction of additional uncertainty has a major effect on the optimal actions, which the agent was able to adapt to adequately.},
note = {Lopes Dos Santos, Bruno (mentor); Yorke-Smith, Neil (graduation committee); de Croon, Guido (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Current state-of-the-art airline planning models are required to decrease models either in size or complexity due to computational limitations, limiting the <br/>operational applicability to problems of representative sizes. Models return suboptimal solutions, especially when confronted with factors of uncertainty. Considering the growing interest in the application of Machine Learning techniques in the Operations Research domain, and the proven success in other fields such as robotics, this research investigates the applicability of these techniques for airline planning. An Advantage Actor-Critic (A2C) Reinforcement Learning agent is applied to the airline fleet planning problem. Because of the increased computational efficiency of using an A2C agent, the problem is increased in size and the highly volatile uncertainty in fuel price is implemented.<br/>Conversion was achieved, and when evaluating the quality of the solutions compared to a deterministic model, the performance was very satisfactory. The A2C agent was able to outperform the deterministic model, with an increasing performance as more complexity was added to the problem. It was found that<br/>the introduction of additional uncertainty has a major effect on the optimal actions, which the agent was able to adapt to adequately.
Willemsen, Daniël
Sample-efficient multi-agent reinforcement learning using learned world models Masters Thesis
TU Delft Aerospace Engineering, 2021, (Coppola, M. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:e44e5449-519b-4a21-81e3-a96f1bbb2811,
title = {Sample-efficient multi-agent reinforcement learning using learned world models},
author = {Daniël Willemsen},
url = {http://resolver.tudelft.nl/uuid:e44e5449-519b-4a21-81e3-a96f1bbb2811},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Multi-agent robotic systems could benefit from reinforcement learning algorithms that are able to learn behaviours in a small number trials, a property known as sample efficiency. This research investigates the use of learned world models to create more sample-efficient algorithms. We present a novel multi-agent model-based reinforcement learning algorithm: Multi-Agent Model-Based Policy Optimization (MAMBPO), utilizing the Centralized Learning for Decentralized Execution (CLDE) framework, and demonstrate state-of-the-art performance in terms of sample efficiency on a number of benchmark domains. CLDE algorithms allow a group of agents to act in a fully decentralized manner after training. This is a desirable property for many systems comprising of multiple robots. Current CLDE algorithms such as Multi-Agent Soft Actor-Critic (MASAC) suffer from limited sample efficiency, often taking many thousands of trials before learning desirable behaviours. This makes these algorithms impractical for learning in real-world robotic tasks. MAMBPO utilizes a learned world model to improve sample efficiency compared to its model-free counterparts. We demonstrate on two simulated multi-agent robotics tasks that MAMBPO is able to reach similar performance to MASAC with up to 3.7 times fewer samples required for learning. Doing this, we take an important step towards making real-life learning for multi-agent robotic systems possible.},
note = {Coppola, M. (mentor); de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Multi-agent robotic systems could benefit from reinforcement learning algorithms that are able to learn behaviours in a small number trials, a property known as sample efficiency. This research investigates the use of learned world models to create more sample-efficient algorithms. We present a novel multi-agent model-based reinforcement learning algorithm: Multi-Agent Model-Based Policy Optimization (MAMBPO), utilizing the Centralized Learning for Decentralized Execution (CLDE) framework, and demonstrate state-of-the-art performance in terms of sample efficiency on a number of benchmark domains. CLDE algorithms allow a group of agents to act in a fully decentralized manner after training. This is a desirable property for many systems comprising of multiple robots. Current CLDE algorithms such as Multi-Agent Soft Actor-Critic (MASAC) suffer from limited sample efficiency, often taking many thousands of trials before learning desirable behaviours. This makes these algorithms impractical for learning in real-world robotic tasks. MAMBPO utilizes a learned world model to improve sample efficiency compared to its model-free counterparts. We demonstrate on two simulated multi-agent robotics tasks that MAMBPO is able to reach similar performance to MASAC with up to 3.7 times fewer samples required for learning. Doing this, we take an important step towards making real-life learning for multi-agent robotic systems possible.
Kemmeren, Max
Improving the Performance of INDI Flight Control for a Quadrotor in the Ceiling Effect Masters Thesis
TU Delft Aerospace Engineering, 2021, (Smeur, E.J.J. (mentor); de Wagter, C. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:35ad1e1c-a3c3-456a-a899-06dac1dc3398,
title = {Improving the Performance of INDI Flight Control for a Quadrotor in the Ceiling Effect},
author = {Max Kemmeren},
url = {http://resolver.tudelft.nl/uuid:35ad1e1c-a3c3-456a-a899-06dac1dc3398},
year = {2021},
date = {2021-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {As the application areas of Unmanned Aerial Vehicles (UAVs) keep expanding, new flight areas are encountered more often. Small UAVs, named Micro Air Vehicles (MAVs), even fly in areas like sewage pipes. These areas introduce new difficulties such as aerodynamic effects caused by the ground and/or ceiling. In this paper two main contributions are presented that deal with the aerodynamic effects caused by the ceiling: 1) an adaptive model describing the ceiling effect using onboard measurements, which can be altered to describe other aerodynamic effects that occur when flying in constrained spaces, 2) incorporating the adaptive model into an Incremental Nonlinear Dynamic Inversion (INDI) controller. The controller is implemented and tested onto a MAV (Crazyflie). The results have shown stability improvements for close ceiling flight. Moreover the minimal distance the MAV can fly from the ceiling is decreased using the new controller.},
note = {Smeur, E.J.J. (mentor); de Wagter, C. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
As the application areas of Unmanned Aerial Vehicles (UAVs) keep expanding, new flight areas are encountered more often. Small UAVs, named Micro Air Vehicles (MAVs), even fly in areas like sewage pipes. These areas introduce new difficulties such as aerodynamic effects caused by the ground and/or ceiling. In this paper two main contributions are presented that deal with the aerodynamic effects caused by the ceiling: 1) an adaptive model describing the ceiling effect using onboard measurements, which can be altered to describe other aerodynamic effects that occur when flying in constrained spaces, 2) incorporating the adaptive model into an Incremental Nonlinear Dynamic Inversion (INDI) controller. The controller is implemented and tested onto a MAV (Crazyflie). The results have shown stability improvements for close ceiling flight. Moreover the minimal distance the MAV can fly from the ceiling is decreased using the new controller.
2020
da Cruz Rodrigues Lourenço, Eduardo Falcão
An intelligent leader-follower neural controller in adverse observability scenarios Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Coppola, M. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:15f2daf8-1596-4514-8b1b-e139881cfaf3,
title = {An intelligent leader-follower neural controller in adverse observability scenarios},
author = {Eduardo Falcão da Cruz Rodrigues Lourenço},
url = {http://resolver.tudelft.nl/uuid:15f2daf8-1596-4514-8b1b-e139881cfaf3},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {A high-level neural controller for leader-follower flight is presented. State of the art range-based relative localization schemes that rely exclusively on onboard sensors present an additional challenge to the leader-follower control problem since they restrict the flight conditions that guarantee observability. This novel controller was developed over an evolutionary process in which the simulation environment resembled the real-life constraints a group of MAVs would encounter. During the learning stage, a group of three agents is used, where one acts as a leader and flies a random trajectory, and the other two act as followers guided by a candidate controller that dictates the desired velocity commands. In the end, when equipped with the best-evolved controller, the follower agents are able to showcase a successful following behaviour that also enhances the observability of the system, although no observability metric was included in evolution.},
note = {de Croon, G.C.H.E. (mentor); Coppola, M. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
A high-level neural controller for leader-follower flight is presented. State of the art range-based relative localization schemes that rely exclusively on onboard sensors present an additional challenge to the leader-follower control problem since they restrict the flight conditions that guarantee observability. This novel controller was developed over an evolutionary process in which the simulation environment resembled the real-life constraints a group of MAVs would encounter. During the learning stage, a group of three agents is used, where one acts as a leader and flies a random trajectory, and the other two act as followers guided by a candidate controller that dictates the desired velocity commands. In the end, when equipped with the best-evolved controller, the follower agents are able to showcase a successful following behaviour that also enhances the observability of the system, although no observability metric was included in evolution.
Veder, Rano
AvoidBench: A high-fidelity vision-based obstacle avoidance benchmarking suite for multi-rotors Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:302426ff-a4a4-4f8a-967a-331ea71b1ba1,
title = {AvoidBench: A high-fidelity vision-based obstacle avoidance benchmarking suite for multi-rotors},
author = {Rano Veder},
url = {http://resolver.tudelft.nl/uuid:302426ff-a4a4-4f8a-967a-331ea71b1ba1},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Obstacle avoidance is an essential topic in the field of autonomous drone research. When choosing an avoidance algorithm, many different options are available, each with their advantages and disadvantages. As there is currently no consensus on testing methods, it is quite challenging to compare the performance between algorithms. In this paper, we present AvoidBench, a benchmarking suite capable of evaluating the performance of vision-based obstacle avoidance algorithms for multi-rotors in simulation. Utilising a set of performance metrics, AvoidBench assigns performance scores to obstacle avoidance algorithms by subjecting them to a series of tasks. Using both Airsim and Unreal engine under the hood, we are able to provide high-fidelity visuals and dynamics, leading to a relatively small gap between simulation and reality. AvoidBench comes included with a simple, but powerful C++ and Python API which provides functionality for procedural environment generation, custom benchmark design, and an easy-to-use framework for users to implement their own vision-based obstacle avoidance methods. Implementing an obstacle avoidance method can be done entirely in a single file, allowing anyone to share and compare their obstacle detection and avoidance algorithms with others.},
note = {de Croon, G.C.H.E. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Obstacle avoidance is an essential topic in the field of autonomous drone research. When choosing an avoidance algorithm, many different options are available, each with their advantages and disadvantages. As there is currently no consensus on testing methods, it is quite challenging to compare the performance between algorithms. In this paper, we present AvoidBench, a benchmarking suite capable of evaluating the performance of vision-based obstacle avoidance algorithms for multi-rotors in simulation. Utilising a set of performance metrics, AvoidBench assigns performance scores to obstacle avoidance algorithms by subjecting them to a series of tasks. Using both Airsim and Unreal engine under the hood, we are able to provide high-fidelity visuals and dynamics, leading to a relatively small gap between simulation and reality. AvoidBench comes included with a simple, but powerful C++ and Python API which provides functionality for procedural environment generation, custom benchmark design, and an easy-to-use framework for users to implement their own vision-based obstacle avoidance methods. Implementing an obstacle avoidance method can be done entirely in a single file, allowing anyone to share and compare their obstacle detection and avoidance algorithms with others.
Duisterhof, Bart
Sniffy Bug: A fully autonomous and collaborative swarm of gas-seeking nano quadcopters in cluttered environments Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Verhoeven, C.J.M. (graduation committee); de Wagter, C. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:7dd7edc9-0037-4c3e-a667-aac7476f272f,
title = {Sniffy Bug: A fully autonomous and collaborative swarm of gas-seeking nano quadcopters in cluttered environments},
author = {Bart Duisterhof},
url = {http://resolver.tudelft.nl/uuid:7dd7edc9-0037-4c3e-a667-aac7476f272f},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Nano quadcopters are ideal for gas source localization (GSL) as they are cheap, safe and agile. However, previous algorithms are unsuitable for nano quadcopters, as they rely on heavy sensors, require too large computational resources, or only solve simple scenarios without obstacles. In this work, we propose a novel bug algorithm named `Sniffy Bug', that allows a swarm of gas-seeking nano quadcopters to localize a gas source in an unknown, cluttered and GPS-denied environment. Sniffy Bug is capable of efficient GSL with extremely little sensory input and computational resources, operating within the strict resource constraints of a nano quadcopter. The algorithm foresees in the avoidance of obstacles and other swarm members, while pursuing desired waypoints. The waypoints are set for exploration, and, when a single swarm member has sensed the gas, by a particle swarm optimization-based procedure. We evolve all the parameters of the bug (and PSO) algorithm, with a novel automated end-to-end simulation and benchmark platform, AutoGDM. This platform enables fully automated end-to-end environment generation and gas dispersion modelling (GDM), not only allowing for learning in simulation but also providing the first GSL benchmark. We show that evolved Sniffy Bug outperforms manually selected parameters in challenging, cluttered environments in the real world. To this end, we show that a lightweight and mapless bug algorithm can be evolved to complete a complex task, and enable the first fully autonomous swarm of collaborative gas-seeking nano quadcopters.},
note = {de Croon, G.C.H.E. (mentor); Verhoeven, C.J.M. (graduation committee); de Wagter, C. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Nano quadcopters are ideal for gas source localization (GSL) as they are cheap, safe and agile. However, previous algorithms are unsuitable for nano quadcopters, as they rely on heavy sensors, require too large computational resources, or only solve simple scenarios without obstacles. In this work, we propose a novel bug algorithm named `Sniffy Bug', that allows a swarm of gas-seeking nano quadcopters to localize a gas source in an unknown, cluttered and GPS-denied environment. Sniffy Bug is capable of efficient GSL with extremely little sensory input and computational resources, operating within the strict resource constraints of a nano quadcopter. The algorithm foresees in the avoidance of obstacles and other swarm members, while pursuing desired waypoints. The waypoints are set for exploration, and, when a single swarm member has sensed the gas, by a particle swarm optimization-based procedure. We evolve all the parameters of the bug (and PSO) algorithm, with a novel automated end-to-end simulation and benchmark platform, AutoGDM. This platform enables fully automated end-to-end environment generation and gas dispersion modelling (GDM), not only allowing for learning in simulation but also providing the first GSL benchmark. We show that evolved Sniffy Bug outperforms manually selected parameters in challenging, cluttered environments in the real world. To this end, we show that a lightweight and mapless bug algorithm can be evolved to complete a complex task, and enable the first fully autonomous swarm of collaborative gas-seeking nano quadcopters.
Bains, Karan
System Identification of the Delfly Nimble: Modeling of the Lateral Body Dynamics Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Visser, C.C. (mentor); Olejnik, D.A. (mentor); Karasek, M. (mentor); Armanini, S.F. (mentor); de Croon, G.C.H.E. (graduation committee); Mooij, E. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:ada45454-c19a-4fba-b842-62efd2320a6a,
title = {System Identification of the Delfly Nimble: Modeling of the Lateral Body Dynamics},
author = {Karan Bains},
url = {http://resolver.tudelft.nl/uuid:ada45454-c19a-4fba-b842-62efd2320a6a},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Flapping wing micro air vehicles (FWMAV's) are a subcategory of unmanned aerial vehicle which use flapping wings for thrust generation. The high agility and maneuverability of FWMAV's are very favorable attributes, making them more applicable in cluttered spaces. A tailless FWMAV called the Delfly Nimble has been developed at the Delft University of Technology. Due to the inherent instability of the tailless design an active controller is required to ensure safe and stable flight of the drone. In previous research, models have been developed for the longitudinal dynamics of the Delfly Nimble. In this paper, a grey-box state-space model of the lateral body dynamics in hover conditions is identified using system identification techniques. The parameters which needed to be estimated were stability and control derivatives, and they were obtained with a least-squares approach. Free-flight experiments were performed to generate the identification and validation data. A doublet train was used in the identification experiments, with the gains of the controller adjusted in such a way that maximum excitation was acquired. The identified model has been validated with various maneuvers. These included doublets, 112-maneuvers, maneuvers using coupled inputs, and maneuvers with sideways flight. The resulting model is able to predict the state derivatives of most maneuver accurately, reaching accuracies of over 90% for maneuvers close to hover. Moreover, in closed-loop configuration it is able to simulate the state response accurately, with accuracies of over 85% for maneuvers close to hover, and remains stable, making it applicable for controller design and stability analysis. Finally, based on the model the inherent instability of the lateral body dynamics was also confirmed, for there are eigenvalues with positive real parts.},
note = {de Visser, C.C. (mentor); Olejnik, D.A. (mentor); Karasek, M. (mentor); Armanini, S.F. (mentor); de Croon, G.C.H.E. (graduation committee); Mooij, E. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Flapping wing micro air vehicles (FWMAV's) are a subcategory of unmanned aerial vehicle which use flapping wings for thrust generation. The high agility and maneuverability of FWMAV's are very favorable attributes, making them more applicable in cluttered spaces. A tailless FWMAV called the Delfly Nimble has been developed at the Delft University of Technology. Due to the inherent instability of the tailless design an active controller is required to ensure safe and stable flight of the drone. In previous research, models have been developed for the longitudinal dynamics of the Delfly Nimble. In this paper, a grey-box state-space model of the lateral body dynamics in hover conditions is identified using system identification techniques. The parameters which needed to be estimated were stability and control derivatives, and they were obtained with a least-squares approach. Free-flight experiments were performed to generate the identification and validation data. A doublet train was used in the identification experiments, with the gains of the controller adjusted in such a way that maximum excitation was acquired. The identified model has been validated with various maneuvers. These included doublets, 112-maneuvers, maneuvers using coupled inputs, and maneuvers with sideways flight. The resulting model is able to predict the state derivatives of most maneuver accurately, reaching accuracies of over 90% for maneuvers close to hover. Moreover, in closed-loop configuration it is able to simulate the state response accurately, with accuracies of over 85% for maneuvers close to hover, and remains stable, making it applicable for controller design and stability analysis. Finally, based on the model the inherent instability of the lateral body dynamics was also confirmed, for there are eigenvalues with positive real parts.
Dellemann, Lars
Hybrid UAV Attitude Control using INDI and Dynamic Tilt-Twist Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Wagter, C. (mentor); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:b2db7412-74d9-4914-94b7-0b922a061adc,
title = {Hybrid UAV Attitude Control using INDI and Dynamic Tilt-Twist},
author = {Lars Dellemann},
url = {http://resolver.tudelft.nl/uuid:b2db7412-74d9-4914-94b7-0b922a061adc},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {The application of Unmanned Aerial Vehicles (UAVs) is increasing, much like the performance of these aircraft. A tailsitter is a type of UAV which is capable of performing vertical take-offs and landings (VTOL) and long endurance flights. During hover, the yaw control is limited due to the dynamics of these tailsitters. The generally used quaternion feedback for the attitude does not compensate for this as it describes a singular rotation. Tilt-twist is a solution to the problem as it splits the tilt (pitch and roll) from the twist (yaw). The axis of the yaw rotation is body fixed. When hovering with a pitch and/or roll angle the twist axis will be aligned with the body z-axis, instead of the desired gravitational force vector (for position control). Previous tilt-twist methods used a PID controller. This paper describes an improvement over previous tilt-twist approaches, the dynamic tilt-twist in combination with INDI. The INDI controller is designed for nonlinear systems. The dynamic tilt-twist compensates for the problem with the normal tilt-twist as test results will demonstrate. Tests are performed in a simulation and a real life test with the NederDrone hybrid tailsitter is done.},
note = {de Wagter, C. (mentor); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
The application of Unmanned Aerial Vehicles (UAVs) is increasing, much like the performance of these aircraft. A tailsitter is a type of UAV which is capable of performing vertical take-offs and landings (VTOL) and long endurance flights. During hover, the yaw control is limited due to the dynamics of these tailsitters. The generally used quaternion feedback for the attitude does not compensate for this as it describes a singular rotation. Tilt-twist is a solution to the problem as it splits the tilt (pitch and roll) from the twist (yaw). The axis of the yaw rotation is body fixed. When hovering with a pitch and/or roll angle the twist axis will be aligned with the body z-axis, instead of the desired gravitational force vector (for position control). Previous tilt-twist methods used a PID controller. This paper describes an improvement over previous tilt-twist approaches, the dynamic tilt-twist in combination with INDI. The INDI controller is designed for nonlinear systems. The dynamic tilt-twist compensates for the problem with the normal tilt-twist as test results will demonstrate. Tests are performed in a simulation and a real life test with the NederDrone hybrid tailsitter is done.
Mattar, Avinash
Acoustic Perception in Intelligent Vehicles using a single microphone system Masters Thesis
TU Delft Mechanical, Maritime and Materials Engineering; TU Delft Cognitive Robotics; TU Delft Intelligent Vehicles, 2020, (Kooij, J.F.P. (mentor); Hehn, T.M. (mentor); Gavrila, D. (graduation committee); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:054d38f7-8fe7-447c-b9fc-9d13437fcef5,
title = {Acoustic Perception in Intelligent Vehicles using a single microphone system},
author = {Avinash Mattar},
url = {http://resolver.tudelft.nl/uuid:054d38f7-8fe7-447c-b9fc-9d13437fcef5},
year = {2020},
date = {2020-01-01},
school = {TU Delft Mechanical, Maritime and Materials Engineering; TU Delft Cognitive Robotics; TU Delft Intelligent Vehicles},
abstract = {Passive acoustic sensing utilizes the ability of sound to travel beyond the line-of-sight to understand the surroundings. This provides an advantage over the currently used sensors in Intelligent Vehicles that can sense obstacles within their line-of-sight only. Recently, a localization based approach has been implemented to take advantage of this sensing modality to predict approaching vehicles behind the blind corner in an urban scenario. While this approach shows a lot of promise, there is a difficulty in integrating the multi-microphone system. Additionally, the system would be unable to differentiate between the nature of two sound sources. This motivates the exploration of a classification based approach which uses audio data from only a single microphone to identify the sound sources present in them. This thesis investigates the possibility of having such a system on the Intelligent Vehicle to predict approaching vehicles from behind the blind corners. A review of the literature revealed that techniques categorized under Sound Event Detection (SED) are suitable to implement a classification based approach. The prediction of the vehicle is treated as a binary classification problem and a Convolutional Recurrent Neural Network (CRNN) is used as the acoustic model to detect the presence of an approaching car in the audio sample represented by Log Mel Spectrogram features. Additionally, domain adaptation techniques were implemented to explore the possibility<br/>of improving the system performance with limited data collected while the ego-vehicle is driving. Experiments carried out indicate that when the ego-vehicle is static, the system performs well with the approaching vehicle predicted 1.4s before it is in line-of-sight and a balanced accuracy of 86.9% achieved for the classification task. However, the system achieved an accuracy of 68% on the samples recorded while the ego-vehicle was driving. Further experiments indicate that the acoustic model cannot generalize well to unseen situations in most cases and experiment with domain adaptation did not show<br/>any improvement in performance.},
note = {Kooij, J.F.P. (mentor); Hehn, T.M. (mentor); Gavrila, D. (graduation committee); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Passive acoustic sensing utilizes the ability of sound to travel beyond the line-of-sight to understand the surroundings. This provides an advantage over the currently used sensors in Intelligent Vehicles that can sense obstacles within their line-of-sight only. Recently, a localization based approach has been implemented to take advantage of this sensing modality to predict approaching vehicles behind the blind corner in an urban scenario. While this approach shows a lot of promise, there is a difficulty in integrating the multi-microphone system. Additionally, the system would be unable to differentiate between the nature of two sound sources. This motivates the exploration of a classification based approach which uses audio data from only a single microphone to identify the sound sources present in them. This thesis investigates the possibility of having such a system on the Intelligent Vehicle to predict approaching vehicles from behind the blind corners. A review of the literature revealed that techniques categorized under Sound Event Detection (SED) are suitable to implement a classification based approach. The prediction of the vehicle is treated as a binary classification problem and a Convolutional Recurrent Neural Network (CRNN) is used as the acoustic model to detect the presence of an approaching car in the audio sample represented by Log Mel Spectrogram features. Additionally, domain adaptation techniques were implemented to explore the possibility<br/>of improving the system performance with limited data collected while the ego-vehicle is driving. Experiments carried out indicate that when the ego-vehicle is static, the system performs well with the approaching vehicle predicted 1.4s before it is in line-of-sight and a balanced accuracy of 86.9% achieved for the classification task. However, the system achieved an accuracy of 68% on the samples recorded while the ego-vehicle was driving. Further experiments indicate that the acoustic model cannot generalize well to unseen situations in most cases and experiment with domain adaptation did not show<br/>any improvement in performance.
Patel, Nishant
Dynamic Modelling and State Estimation of a High Speed Racing Drone Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Xu, Y. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:a88b7802-2b7c-44cb-85f6-63d0453fc9e7,
title = {Dynamic Modelling and State Estimation of a High Speed Racing Drone},
author = {Nishant Patel},
url = {http://resolver.tudelft.nl/uuid:a88b7802-2b7c-44cb-85f6-63d0453fc9e7},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Autonomous drone racing has taken a turn for the better in recent years. Drones are becoming faster and implementing better state-of-the-art control techniques to overcome different challenges. With advancements in the fields of computer vision, machine learning, and artificial intelligence, the final goal of autonomous drones is to be quicker than human-piloted racing drones. Increasing the speed of autonomous drones increases the risks associated with flying them. Time-optimal control algorithms have been identified as a method of implementing<br/>aggressive maneuvers to fly drones at high speeds throughout the course of the race. These methods require precise state-estimates. This research work identifies a model for the rate controller. The work also includes an implementation of a state estimation model with drag compensation, also merging a pre-existing refined thrust model with Coriolis effects. With the idea of developing a state estimation model for a racing drone, the model is improved to<br/>include flight envelopes involving motor saturations.},
note = {de Croon, G.C.H.E. (mentor); Xu, Y. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Autonomous drone racing has taken a turn for the better in recent years. Drones are becoming faster and implementing better state-of-the-art control techniques to overcome different challenges. With advancements in the fields of computer vision, machine learning, and artificial intelligence, the final goal of autonomous drones is to be quicker than human-piloted racing drones. Increasing the speed of autonomous drones increases the risks associated with flying them. Time-optimal control algorithms have been identified as a method of implementing<br/>aggressive maneuvers to fly drones at high speeds throughout the course of the race. These methods require precise state-estimates. This research work identifies a model for the rate controller. The work also includes an implementation of a state estimation model with drag compensation, also merging a pre-existing refined thrust model with Coriolis effects. With the idea of developing a state estimation model for a racing drone, the model is improved to<br/>include flight envelopes involving motor saturations.
Wijngaarden, Dennis
Implicit Coordinated Tactical Avoidance for UAVs within a Geofenced Airspace Masters Thesis
TU Delft Aerospace Engineering, 2020, (Ellerbroek, Joost (mentor); Remes, B.D.W. (mentor); Hoekstra, J.M. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:4b92f6b0-dc40-4946-a1ae-7efd0df79401,
title = {Implicit Coordinated Tactical Avoidance for UAVs within a Geofenced Airspace},
author = {Dennis Wijngaarden},
url = {http://resolver.tudelft.nl/uuid:4b92f6b0-dc40-4946-a1ae-7efd0df79401},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {This research presents the derivation, implementation and safety assessment of a velocity obstacle- based conflict resolution method to be used by UAVs flying within a horizontally restricted airspace by a geofence under the presence of wind. Two parameters indicating the safety of the applied conflict resolution method have been measured, i.e., the Intrusion Prevention Rate (IPR) and the Violation Prevention Rate of the Geofence (VPRG). Three coordination rule-sets have been implemented i.e., 1) geometric optimum (OPT), 2) geometric optimum from target heading (DEST) and 3) only change in heading (HDG). These rule-sets have been assessed during a safety assessment. It was concluded that the OPT rule-set performed best in terms of the IPR and the DEST rule-set performed best in terms of the VPRG under windy and wind calm conditions. The HDG rule-set performed worst in terms of both safety parameters. It was noted that both safety parameters are the lowest when conflicts occur close the geofence under windy conditions for all implemented rule-sets.},
note = {Ellerbroek, Joost (mentor); Remes, B.D.W. (mentor); Hoekstra, J.M. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
This research presents the derivation, implementation and safety assessment of a velocity obstacle- based conflict resolution method to be used by UAVs flying within a horizontally restricted airspace by a geofence under the presence of wind. Two parameters indicating the safety of the applied conflict resolution method have been measured, i.e., the Intrusion Prevention Rate (IPR) and the Violation Prevention Rate of the Geofence (VPRG). Three coordination rule-sets have been implemented i.e., 1) geometric optimum (OPT), 2) geometric optimum from target heading (DEST) and 3) only change in heading (HDG). These rule-sets have been assessed during a safety assessment. It was concluded that the OPT rule-set performed best in terms of the IPR and the DEST rule-set performed best in terms of the VPRG under windy and wind calm conditions. The HDG rule-set performed worst in terms of both safety parameters. It was noted that both safety parameters are the lowest when conflicts occur close the geofence under windy conditions for all implemented rule-sets.
Karssies, Jan
Extended Incremental Non-linear Control Allocation on the TU Delft Quadplane Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Wagter, C. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:71257d1e-c65b-4eb7-9df0-869b9419a8c2,
title = {Extended Incremental Non-linear Control Allocation on the TU Delft Quadplane},
author = {Jan Karssies},
url = {http://resolver.tudelft.nl/uuid:71257d1e-c65b-4eb7-9df0-869b9419a8c2},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {This research presents an implementation of a novel controller design on an overactuated hybrid Unmanned Aerial Vehicle (UAV). This platform is a hybrid between a conventional quadcopter and a fixed-wing aircraft. Its inner loop is controlled by an existing but modified control method called Incremental Non-linear Control Allocation or INCA. This controller deals with the platform’s control allocation problem by minimising a set of objective functions with a method known as the Active Set Method and avoids actuator saturation. For the vehicle’s outer loop, a novel extension to INCA is presented, called Extended INCA or XINCA. This method optimises one of the physical actuator’s command and the angular control demands fed to the vehicle’s inner loop, based on linear reference accelerations. It does so while adapting to varying flight phases, conditions and vehicle states, and taking into account the aerodynamic properties of the main wing. XINCA has low dependence on accurate vehicle models and requires configuration using only several optimisation parameters. Both flight simulations and experimental flights are performed to prove the performance of both controllers.},
note = {de Wagter, C. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
This research presents an implementation of a novel controller design on an overactuated hybrid Unmanned Aerial Vehicle (UAV). This platform is a hybrid between a conventional quadcopter and a fixed-wing aircraft. Its inner loop is controlled by an existing but modified control method called Incremental Non-linear Control Allocation or INCA. This controller deals with the platform’s control allocation problem by minimising a set of objective functions with a method known as the Active Set Method and avoids actuator saturation. For the vehicle’s outer loop, a novel extension to INCA is presented, called Extended INCA or XINCA. This method optimises one of the physical actuator’s command and the angular control demands fed to the vehicle’s inner loop, based on linear reference accelerations. It does so while adapting to varying flight phases, conditions and vehicle states, and taking into account the aerodynamic properties of the main wing. XINCA has low dependence on accurate vehicle models and requires configuration using only several optimisation parameters. Both flight simulations and experimental flights are performed to prove the performance of both controllers.
Cheung, Louis
Fuel Cell Drone for Soil Monitoring Masters Thesis
TU Delft Applied Sciences; TU Delft Electrical Engineering, Mathematics and Computer Science, 2020, (Aravind, P.V. (mentor); Bhattacharya, Nandini (graduation committee); Remes, Bart (graduation committee); Tambi, Yash (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:a191b5fc-4050-44b8-9652-4493d44b654c,
title = {Fuel Cell Drone for Soil Monitoring},
author = {Louis Cheung},
url = {http://resolver.tudelft.nl/uuid:a191b5fc-4050-44b8-9652-4493d44b654c},
year = {2020},
date = {2020-01-01},
school = {TU Delft Applied Sciences; TU Delft Electrical Engineering, Mathematics and Computer Science},
abstract = {We have set out to develop a drone, based on the existing Delftacopter, capable of soil monitoring via LiDAR remote sensing. The battery was to be replaced by a fuel cell system in order to extend the range threefold to 180km. Unfortunately, the ultimate design is likely unfeasible. Agriculture requires healthy soil and monitoring soil health is fundamental to its maintenance. Soil organic carbon in particular provides energy to the soil’s microorganisms, and is beneficial to water and nutrient retention. In addition, storing carbon in the soil is a form of carbon sequestration, which has become interesting due to the rising levels of carbon dioxide in our atmosphere. Monitoring soil organic carbon is therefore the goal of the drone design. The fuel cell system is a 650Whydrogen fuel cell by Intelligent Energy with a mass of 1290 g, which will be replacing the battery in the base design. Fuel tanks that were considered suitable are the 450 g, 0.5 L, 500 bar and 1350 g, 3 L, 300 bar fuel tanks by Meyer. It was found that one 1350 g and two 450 g fuel tanks were necessary to achieve the desired range of 180 km. However, after more careful drag estimates, this configuration turns out to be too heavy. 4 450 g fuel tanks remains feasible. Results below are based on this amount of fuel tanks. The incorporated LiDAR sensor is one by Velodyne, namely the Puck LITE, with a specified range of 100 m. The LiDAR sensor has a firing cycle of 55.296 &s, almost 20 kHz. Based on previous studies that used LiDAR to measure soil organic carbon, it has been established that a density of 5 data points per square meter is required. Fromour LiDAR parameters it turns out that the optimal flight altitude is 27.5mabove the surface that is to be measured, with a rotation rate of 10 Hz for the LiDAR sensor, when flying at a speed of 20ms¡1. With a flight distance of roughly 116km at 22.5ms¡1 (111km at 20ms¡1), an area of 21.8km2 per flight can be scanned. 1},
note = {Aravind, P.V. (mentor); Bhattacharya, Nandini (graduation committee); Remes, Bart (graduation committee); Tambi, Yash (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
We have set out to develop a drone, based on the existing Delftacopter, capable of soil monitoring via LiDAR remote sensing. The battery was to be replaced by a fuel cell system in order to extend the range threefold to 180km. Unfortunately, the ultimate design is likely unfeasible. Agriculture requires healthy soil and monitoring soil health is fundamental to its maintenance. Soil organic carbon in particular provides energy to the soil’s microorganisms, and is beneficial to water and nutrient retention. In addition, storing carbon in the soil is a form of carbon sequestration, which has become interesting due to the rising levels of carbon dioxide in our atmosphere. Monitoring soil organic carbon is therefore the goal of the drone design. The fuel cell system is a 650Whydrogen fuel cell by Intelligent Energy with a mass of 1290 g, which will be replacing the battery in the base design. Fuel tanks that were considered suitable are the 450 g, 0.5 L, 500 bar and 1350 g, 3 L, 300 bar fuel tanks by Meyer. It was found that one 1350 g and two 450 g fuel tanks were necessary to achieve the desired range of 180 km. However, after more careful drag estimates, this configuration turns out to be too heavy. 4 450 g fuel tanks remains feasible. Results below are based on this amount of fuel tanks. The incorporated LiDAR sensor is one by Velodyne, namely the Puck LITE, with a specified range of 100 m. The LiDAR sensor has a firing cycle of 55.296 &s, almost 20 kHz. Based on previous studies that used LiDAR to measure soil organic carbon, it has been established that a density of 5 data points per square meter is required. Fromour LiDAR parameters it turns out that the optimal flight altitude is 27.5mabove the surface that is to be measured, with a rotation rate of 10 Hz for the LiDAR sensor, when flying at a speed of 20ms¡1. With a flight distance of roughly 116km at 22.5ms¡1 (111km at 20ms¡1), an area of 21.8km2 per flight can be scanned. 1
Büller, Bas
Supervised Learning in Spiking Neural Networks Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, Guido (mentor); Paredes Valles, Federico (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:256f7044-862d-4b53-b395-973dadbb7a00,
title = {Supervised Learning in Spiking Neural Networks},
author = {Bas Büller},
url = {http://resolver.tudelft.nl/uuid:256f7044-862d-4b53-b395-973dadbb7a00},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Spiking neural networks are notoriously hard to train because of their complex dynamics and sparse spiking signals. However, in part due to these properties, spiking neurons possess high computa- tional power and high theoretical energy efficiency. This thesis introduces an online, supervised, and gradient-based learning algorithm for spiking neural networks. It is shown how gradients of temporal signals that influence spiking neurons can be calculated online as an eligibility trace. The trace rep- resents the temporal gradient as a single scalar value and is recursively updated at each consecutive iteration. Moreover, the learning method uses approximate error signals to simplify their calculation and make the error calculation compatible with online learning. In several experiments, it is shown that the algorithm can solve spatial credit assignment problems with short-term temporal dependencies in deep spiking neural networks. Potential approaches for improving the algorithm’s performance on long-term temporal credit assignment problems are also discussed. Besides the research on spiking neural networks, this thesis includes an in-depth literature study on the topics of neuromorphic computing and deep learning, as well as extensive evaluations of several learning algorithms for spiking neural networks},
note = {de Croon, Guido (mentor); Paredes Valles, Federico (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Spiking neural networks are notoriously hard to train because of their complex dynamics and sparse spiking signals. However, in part due to these properties, spiking neurons possess high computa- tional power and high theoretical energy efficiency. This thesis introduces an online, supervised, and gradient-based learning algorithm for spiking neural networks. It is shown how gradients of temporal signals that influence spiking neurons can be calculated online as an eligibility trace. The trace rep- resents the temporal gradient as a single scalar value and is recursively updated at each consecutive iteration. Moreover, the learning method uses approximate error signals to simplify their calculation and make the error calculation compatible with online learning. In several experiments, it is shown that the algorithm can solve spatial credit assignment problems with short-term temporal dependencies in deep spiking neural networks. Potential approaches for improving the algorithm’s performance on long-term temporal credit assignment problems are also discussed. Besides the research on spiking neural networks, this thesis includes an in-depth literature study on the topics of neuromorphic computing and deep learning, as well as extensive evaluations of several learning algorithms for spiking neural networks
Jong, David
How do deep neural networks perform optical flow estimation?: A neuropsychology-inspired study Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Paredes-Vallés, Federico (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:64d40dfc-d852-4688-b8f4-af37f3e9704c,
title = {How do deep neural networks perform optical flow estimation?: A neuropsychology-inspired study},
author = {David Jong},
url = {http://resolver.tudelft.nl/uuid:64d40dfc-d852-4688-b8f4-af37f3e9704c},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {End-to-end trained Convolutional Neural Networks have led to a breakthrough in optical flow estimation. The most recent advances focus on improving the optical flow estimation by improving the architecture and setting a new benchmark on the publicly available MPI-Sintel dataset. Instead, in this article, we investigate how deep neural networks estimate optical flow. By obtaining an understanding of how these networks function, more can be said about the behavior of these networks in unexpected scenarios and how the architecture and training data can be improved to obtain a better performance. For our investigation, we use a filter identification method that has played a major role in uncovering the motion filters present in animal brains in neuropsychological research. The method shows that the filters in deep neural networks are sensitive to a variety of motion patterns. Not only do we find translation filters, as demonstrated in animal brains, but thanks to the easier measurements in artificial neural networks, we even unveil dilation, rotation and occlusion filters. Furthermore, we find similarities in the refinement part of the network and the perceptual filling-in process which occurs in the mammal primary visual cortex.},
note = {de Croon, G.C.H.E. (mentor); Paredes-Vallés, Federico (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
End-to-end trained Convolutional Neural Networks have led to a breakthrough in optical flow estimation. The most recent advances focus on improving the optical flow estimation by improving the architecture and setting a new benchmark on the publicly available MPI-Sintel dataset. Instead, in this article, we investigate how deep neural networks estimate optical flow. By obtaining an understanding of how these networks function, more can be said about the behavior of these networks in unexpected scenarios and how the architecture and training data can be improved to obtain a better performance. For our investigation, we use a filter identification method that has played a major role in uncovering the motion filters present in animal brains in neuropsychological research. The method shows that the filters in deep neural networks are sensitive to a variety of motion patterns. Not only do we find translation filters, as demonstrated in animal brains, but thanks to the easier measurements in artificial neural networks, we even unveil dilation, rotation and occlusion filters. Furthermore, we find similarities in the refinement part of the network and the perceptual filling-in process which occurs in the mammal primary visual cortex.
Spronk, Simon
An NonlinearModel Predictive Control Approach to Autonomous UAV Racing Trajectory Generation & Control Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, Guido (mentor); Li, Shuo (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:88e689a1-eb19-4e48-91b2-3b122a824503,
title = {An NonlinearModel Predictive Control Approach to Autonomous UAV Racing Trajectory Generation & Control},
author = {Simon Spronk},
url = {http://resolver.tudelft.nl/uuid:88e689a1-eb19-4e48-91b2-3b122a824503},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {When observing an Autonomous Unmanned Aerial Vehicle(UAV) race, one would be hard-pressed to call it racing as the actual velocities attained are extremely low. This article addresses this shortcoming by proposing a method of generating and executing a racing trajectory for a UAV, through a series of position objectives representative of a racing environment, with the goal of significantly improving the velocity when compared to the current norm of PID controllers. The method consists of applying Nonlinear Model Predictive Control with the capability of dynamically updating the position goal based upon internal state estimation to generate a set of inputs for a UAV. To prove the viability of the proposed method we test by using numerical simulations, a flight simulator environment(Gazebo) and a series of real-world flight tests on the Bebop1 UAV. Through 2 iterations of the testing process it is proven that the method is able to significantly decrease the flight time ()through both simple and more complex short range manoeuvres(2m-4m). However model errors and an inability to fully control thrust on the UAV introduce a significant and consistent position error.},
note = {de Croon, Guido (mentor); Li, Shuo (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
When observing an Autonomous Unmanned Aerial Vehicle(UAV) race, one would be hard-pressed to call it racing as the actual velocities attained are extremely low. This article addresses this shortcoming by proposing a method of generating and executing a racing trajectory for a UAV, through a series of position objectives representative of a racing environment, with the goal of significantly improving the velocity when compared to the current norm of PID controllers. The method consists of applying Nonlinear Model Predictive Control with the capability of dynamically updating the position goal based upon internal state estimation to generate a set of inputs for a UAV. To prove the viability of the proposed method we test by using numerical simulations, a flight simulator environment(Gazebo) and a series of real-world flight tests on the Bebop1 UAV. Through 2 iterations of the testing process it is proven that the method is able to significantly decrease the flight time ()through both simple and more complex short range manoeuvres(2m-4m). However model errors and an inability to fully control thrust on the UAV introduce a significant and consistent position error.
Spronk, Simon
An Nonlinear Model Predictive Control Approach to Autonomous UAV Racing Trajectory Generation and Control Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Li, S. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:fc2e13cb-4ea1-4aa7-b7f2-1d8d9478daf4,
title = {An Nonlinear Model Predictive Control Approach to Autonomous UAV Racing Trajectory Generation and Control},
author = {Simon Spronk},
url = {http://resolver.tudelft.nl/uuid:fc2e13cb-4ea1-4aa7-b7f2-1d8d9478daf4},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {hen observing an Autonomous Unmanned Aerial Vehicle(UAV) race, one would be hard-pressed to call it racing as the actual velocities attained are extremely low. This article addresses this shortcoming by proposing a method of generating and executing a racing trajectory for a UAV, through a series of position objectives representative of a racing environment, with the goal of significantly improving the velocity when compared to the current norm of PID controllers. The method consists of applying Nonlinear Model Predictive Control with the capability of dynamically updating the position goal based upon internal state estimation to generate a set of inputs for a UAV. To prove the viability of the proposed method we test by using numerical simulations, a flight simulator environment(Gazebo) and a series of real-world flight tests on the Bebop1 UAV. Through 2 iterations of the testing process it is proven that the method is able to significantly(approximately 1s) decrease the flight time through both simple and more complex short range manoeuvres(2m-4m). However model errors and an inability to fully control thrust on the UAV introduce a significant and consistent position error.},
note = {de Croon, G.C.H.E. (mentor); Li, S. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
hen observing an Autonomous Unmanned Aerial Vehicle(UAV) race, one would be hard-pressed to call it racing as the actual velocities attained are extremely low. This article addresses this shortcoming by proposing a method of generating and executing a racing trajectory for a UAV, through a series of position objectives representative of a racing environment, with the goal of significantly improving the velocity when compared to the current norm of PID controllers. The method consists of applying Nonlinear Model Predictive Control with the capability of dynamically updating the position goal based upon internal state estimation to generate a set of inputs for a UAV. To prove the viability of the proposed method we test by using numerical simulations, a flight simulator environment(Gazebo) and a series of real-world flight tests on the Bebop1 UAV. Through 2 iterations of the testing process it is proven that the method is able to significantly(approximately 1s) decrease the flight time through both simple and more complex short range manoeuvres(2m-4m). However model errors and an inability to fully control thrust on the UAV introduce a significant and consistent position error.
Jong, C. P. L.
Never Landing Drone Masters Thesis
TU Delft Aerospace Engineering, 2020, (Remes, B.D.W. (mentor); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:32a90402-a577-4383-afe3-f8a865a287dc,
title = {Never Landing Drone},
author = {C. P. L. Jong},
url = {http://resolver.tudelft.nl/uuid:32a90402-a577-4383-afe3-f8a865a287dc},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Increasing endurance is a major challenge for battery-powered aerial vehicles. A method is presented which makes use of an updraft around obstacles to decrease the power consumption of a fixed-wing, unmanned aerial vehicle. Simulatory results have shown the conditions that the flight controller can fly in.<br/>The effect of a change in wind velocity, wind direction and updraft has been analysed. The simulations showed that an increase in either updraft or absolute wind direction decrease the throttle consumption.<br/>A change in wind velocity results in a shift of the flight controller’s boundaries. The simulations achieved sustained flight at 0 per cent throttle. The practical, autonomous tests reduced the average throttle down to 4.5 per cent in front of the boat. The unfavourable wind conditions and inaccuracies explain this minor<br/>throttle requirement during the final experiment.},
note = {Remes, B.D.W. (mentor); de Croon, G.C.H.E. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Increasing endurance is a major challenge for battery-powered aerial vehicles. A method is presented which makes use of an updraft around obstacles to decrease the power consumption of a fixed-wing, unmanned aerial vehicle. Simulatory results have shown the conditions that the flight controller can fly in.<br/>The effect of a change in wind velocity, wind direction and updraft has been analysed. The simulations showed that an increase in either updraft or absolute wind direction decrease the throttle consumption.<br/>A change in wind velocity results in a shift of the flight controller’s boundaries. The simulations achieved sustained flight at 0 per cent throttle. The practical, autonomous tests reduced the average throttle down to 4.5 per cent in front of the boat. The unfavourable wind conditions and inaccuracies explain this minor<br/>throttle requirement during the final experiment.
Hagenaars, Jesse
Evolved Neuromorphic Control for High Speed Divergence-based Landings of MAVs Masters Thesis
TU Delft Aerospace Engineering, 2020, (de Croon, G.C.H.E. (mentor); Paredes-Vallés, F. (mentor); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:48040e88-f507-4676-a5da-2b701a07f387,
title = {Evolved Neuromorphic Control for High Speed Divergence-based Landings of MAVs},
author = {Jesse Hagenaars},
url = {http://resolver.tudelft.nl/uuid:48040e88-f507-4676-a5da-2b701a07f387},
year = {2020},
date = {2020-01-01},
school = {TU Delft Aerospace Engineering},
abstract = {Flying insects are capable of autonomous vision-based navigation in cluttered environments, reliably avoiding objects through fast and agile manoeuvres. Meanwhile, insect-scale micro air vehicles still lag far behind their biological counterparts, displaying inferior performance at a fraction of the energy efficiency. In light of this, it is in our interest to try and mimic flying insects in terms of their vision-based navigation capabilities, and consequently apply gained knowledge to a manoeuvre of relevance. This thesis does so through evolving spiking neural networks for controlling divergence-based landings of micro air vehicles, while minimising the network's spike rate. We demonstrate vision-based neuromorphic control for a real-world, continuous problem, as well as the feasibility of extending this controller to one that is end-to-end-learnt, and can work with an event-based camera. Furthermore, we provide insight into the resources required for successfully solving the problem of divergence-based landing, showing that high-resolution control can be learnt with only a single spiking neuron. Finally, we look at evolving only a subset of the spiking neural network's available hyperparameters, suggesting that the best results are obtained when all parameters are affected by the learning process.},
note = {de Croon, G.C.H.E. (mentor); Paredes-Vallés, F. (mentor); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Flying insects are capable of autonomous vision-based navigation in cluttered environments, reliably avoiding objects through fast and agile manoeuvres. Meanwhile, insect-scale micro air vehicles still lag far behind their biological counterparts, displaying inferior performance at a fraction of the energy efficiency. In light of this, it is in our interest to try and mimic flying insects in terms of their vision-based navigation capabilities, and consequently apply gained knowledge to a manoeuvre of relevance. This thesis does so through evolving spiking neural networks for controlling divergence-based landings of micro air vehicles, while minimising the network's spike rate. We demonstrate vision-based neuromorphic control for a real-world, continuous problem, as well as the feasibility of extending this controller to one that is end-to-end-learnt, and can work with an event-based camera. Furthermore, we provide insight into the resources required for successfully solving the problem of divergence-based landing, showing that high-resolution control can be learnt with only a single spiking neuron. Finally, we look at evolving only a subset of the spiking neural network's available hyperparameters, suggesting that the best results are obtained when all parameters are affected by the learning process.
Hagenaars, Jesse
Evolved Neuromorphic Control for High Speed Divergence-based Landings of MAVs Masters Thesis
Delft University of Technology, Delft, NL, 2020.
@mastersthesis{Hagenaars2020,
title = {Evolved Neuromorphic Control for High Speed Divergence-based Landings of MAVs},
author = {Jesse Hagenaars},
url = {http://resolver.tudelft.nl/uuid:48040e88-f507-4676-a5da-2b701a07f387},
year = {2020},
date = {2020-01-01},
address = {Delft, NL},
school = {Delft University of Technology},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
2019
Lyrakis, Alex
Low-cost stereo-based obstacle avoidance for small UAVs using uncertainty maps Masters Thesis
TU Delft Electrical Engineering, Mathematics and Computer Science, 2019, (van Dijk, Tom (mentor); de Croon, G.C.H.E. (mentor); Wong, J.S.S.M. (mentor); Verhoeven, C.J.M. (graduation committee); van Genderen, A.J. (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:96a87f24-c9b1-4a3e-99df-09dc27772609,
title = {Low-cost stereo-based obstacle avoidance for small UAVs using uncertainty maps},
author = {Alex Lyrakis},
url = {http://resolver.tudelft.nl/uuid:96a87f24-c9b1-4a3e-99df-09dc27772609},
year = {2019},
date = {2019-01-01},
school = {TU Delft Electrical Engineering, Mathematics and Computer Science},
abstract = {The last years there is a wide interest in UAVs which can be attributed to their low cost and wide range of use in recreational, commercial and scientific applications. Despite the large increase in drones, UAV flights are permitted only in secluded areas. In order to be granted access to public areas, it must prove its capacity to sense and safely avoid collisions with other obstacles. Therefore, the need for a secure and reliable CAS is imperative. In this thesis a<br/>low-cost, low computationally demanding, stereo-based, robust CAS solution for small UAVs is designed, assuming flights primarily in an outdoor environment. In order to address this problem, firstly the existing dense stereo open-source algorithms are reviewed based on their suitability for obstacle avoidance and their computational complexity. Based on a semantic evaluation and profiling, the review concludes that BM should be preferred for low-cost obstacle avoidance and SGBM should be used only in highly textureless environments. Subsequently, since the imperfect accuracy of any existing stereo solution is a fact, a machine learning method is introduced in order to predict the uncertainty of the stereo measurements. This so called “uncertainty map” method assigns an uncertainty value to every image pixel. It was shown that it can successfully predict the uncertainties of BM and SGBM stereo algorithms. Furthermore, a low-cost collision avoidance method was proposed which makes use of uncertainty map in sensing filtering, collision detection and path planning. The evaluation showed that the use of uncertainty map improves<br/>both collision detection and path planning, especially when BM is used. Last but not least, the whole CAS was implemented in an embedded system Raspberry Pi 3 model B+. Results show that real-time execution of the propsoed CAS with a runtime frequency of 4-54 Hz is possible when BM is used and 1-16 Hz when SGBM is used.},
note = {van Dijk, Tom (mentor); de Croon, G.C.H.E. (mentor); Wong, J.S.S.M. (mentor); Verhoeven, C.J.M. (graduation committee); van Genderen, A.J. (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
The last years there is a wide interest in UAVs which can be attributed to their low cost and wide range of use in recreational, commercial and scientific applications. Despite the large increase in drones, UAV flights are permitted only in secluded areas. In order to be granted access to public areas, it must prove its capacity to sense and safely avoid collisions with other obstacles. Therefore, the need for a secure and reliable CAS is imperative. In this thesis a<br/>low-cost, low computationally demanding, stereo-based, robust CAS solution for small UAVs is designed, assuming flights primarily in an outdoor environment. In order to address this problem, firstly the existing dense stereo open-source algorithms are reviewed based on their suitability for obstacle avoidance and their computational complexity. Based on a semantic evaluation and profiling, the review concludes that BM should be preferred for low-cost obstacle avoidance and SGBM should be used only in highly textureless environments. Subsequently, since the imperfect accuracy of any existing stereo solution is a fact, a machine learning method is introduced in order to predict the uncertainty of the stereo measurements. This so called “uncertainty map” method assigns an uncertainty value to every image pixel. It was shown that it can successfully predict the uncertainties of BM and SGBM stereo algorithms. Furthermore, a low-cost collision avoidance method was proposed which makes use of uncertainty map in sensing filtering, collision detection and path planning. The evaluation showed that the use of uncertainty map improves<br/>both collision detection and path planning, especially when BM is used. Last but not least, the whole CAS was implemented in an embedded system Raspberry Pi 3 model B+. Results show that real-time execution of the propsoed CAS with a runtime frequency of 4-54 Hz is possible when BM is used and 1-16 Hz when SGBM is used.
Stroobants, Stein
On-board Micro Quadrotor State Estimation Using Range Measurements: A Moving Horizon Approach Masters Thesis
TU Delft Mechanical, Maritime and Materials Engineering, 2019, (Keviczky, Tamas (mentor); de Croon, Guido (graduation committee); Li, Shushuai (graduation committee); de Wagter, Christophe (graduation committee); Delft University of Technology (degree granting institution)).
@mastersthesis{uuid:c4170bbd-059d-4874-82f9-958bec2c668e,
title = {On-board Micro Quadrotor State Estimation Using Range Measurements: A Moving Horizon Approach},
author = {Stein Stroobants},
url = {http://resolver.tudelft.nl/uuid:c4170bbd-059d-4874-82f9-958bec2c668e},
year = {2019},
date = {2019-01-01},
school = {TU Delft Mechanical, Maritime and Materials Engineering},
abstract = {Accurate indoor localization is essential for autonomous robotic agents to perform tasks ranging from warehouse management to remote sensing in greenhouses. Recently Ultra Wideband (UWB) distance measurements have been used to estimate position and velocity indoors. These UWB-measurements are known to be corrupted by a varying bias. Besides, current estimation methods are not suitable for large areas with a low beacon coverage. The goal<br/>of this thesis was therefore twofold. First, a simple bias model was proposed to reduce the influence of the UWB bias while still being implementable on a micro-processor. This model was shown to reduce the measurement error with 50% on validation data. Using this model, UWB-localization in a static beacon-configuration can be quickly improved. Second, an adaptation of the standard Moving Horizon Estimation (MHE) method was proposed that uses a time-window of range measurements to increase the robustness to outliers and is still real-time implementable on a micro-processor. This Moving Horizon Model Parametrization (MH-MP) does not estimate every state in the complete time-window, but only estimates an offset of the initial state in the window. An analysis of simulation data and data gathered in flight has shown that the proposed MH-MP outperforms the Extended Kalman Filter (EKF) in both the<br/>position and velocity estimate and has a comparable computation time. Further research is necessary to investigate the possibility of estimating the UWB-bias model parameters online.},
note = {Keviczky, Tamas (mentor); de Croon, Guido (graduation committee); Li, Shushuai (graduation committee); de Wagter, Christophe (graduation committee); Delft University of Technology (degree granting institution)},
keywords = {},
pubstate = {published},
tppubtype = {mastersthesis}
}
Accurate indoor localization is essential for autonomous robotic agents to perform tasks ranging from warehouse management to remote sensing in greenhouses. Recently Ultra Wideband (UWB) distance measurements have been used to estimate position and velocity indoors. These UWB-measurements are known to be corrupted by a varying bias. Besides, current estimation methods are not suitable for large areas with a low beacon coverage. The goal<br/>of this thesis was therefore twofold. First, a simple bias model was proposed to reduce the influence of the UWB bias while still being implementable on a micro-processor. This model was shown to reduce the measurement error with 50% on validation data. Using this model, UWB-localization in a static beacon-configuration can be quickly improved. Second, an adaptation of the standard Moving Horizon Estimation (MHE) method was proposed that uses a time-window of range measurements to increase the robustness to outliers and is still real-time implementable on a micro-processor. This Moving Horizon Model Parametrization (MH-MP) does not estimate every state in the complete time-window, but only estimates an offset of the initial state in the window. An analysis of simulation data and data gathered in flight has shown that the proposed MH-MP outperforms the Extended Kalman Filter (EKF) in both the<br/>position and velocity estimate and has a comparable computation time. Further research is necessary to investigate the possibility of estimating the UWB-bias model parameters online.
B.Sc. Thesis Assignments at TU Delft
Every year the MAVLab also guides one or more 3rd-year DSE Projects (Design and Synthesis Exercises). Students who join this course have the opportunity of being assigned to such projects.
https://mavlab.tudelft.nl/hybrid-mav-wins-1st-prize-dse-symposium/