At ICRA (IEEE International Conference on Robotics and Automation) 2023, MAVLab presented AvoidBench, a high-fidelity vision-based obstacle avoidance benchmarking suite for multi-rotors.
Motivation
Vision-based obstacle avoidance is popular, since vision sensor can be compact, light-weight and low cost. Unfortunately, there is currently no shared consensus on how vision-based obstacle avoidance algorithms should be tested. This makes it not only difficult to see how the field of obstacle avoidance as a whole is progressing, but it also makes it hard to compare the performance of different obstacle avoidance algorithms.
Benchmarking suite based on a high-fidelity simulator
To improve this situation, we have developed “AvoidBench”, a software framework for benchmarking vision-based obstacle avoidance methods on simulated quadrotor drones. The framework allows researchers to test their own avoidance algorithms in different auto-generated environments. The environments are generated to vary on important characteristics that determine the difficulty of the task, including the obstacle density and the amount of visual texture in the environment. The framework employs multiple performance metrics, including among others success rate, path and energy optimality, flight velocity, computational effort, and progress towards the desired waypoint.
The AvoidBench software is available, also as a Docker image, at: https://github.com/tudelft/AvoidBench. Researchers can change both the vision and avoidance algorithms, either in Python or C++.
Sorry, no publications matched your criteria.