H. Fujii, Y. Nagata, I. Ono, and S. Kobayashi (Japan)
Reinforcement Learning, Ecological Psychology, Au tonomous Mobile Robot, τ-margin
In this paper, we introduce τ-margin, which extracts remaining time before collision, into design principle of autonomous mobile robots. We designed and implemented the robot using τ-margin to reduce state space and give the robot robustness of the change in speed. To evaluate the robot, we carried out various experiments based on obstacle avoidance because it is a basic and common task for the autonomous mobile robot. As a result, we confirmed that the robot could achieve obstacle avoidance regardless of the change in its speed and finish learning policies of relatively complicated tasks in a short time. Finally, we verified that the acquired policies have robustness of the speed of robots.
Important Links:
Go Back
