5620 / Step-Climbing Motion Acquisition Of Tracked Robot With Flippers Without Using Environment Information By Reinforcement Learning

Authors

Ryosuke Eto, Hayatake Sato, and Junya Yamakawa

Paper presented at ISTVS 2024 | 21st International and 12th Asia-Pacific Regional Conference of the ISTVS Keywords: tracked robot; flipper; step-climbing; reinforcement learning https://doi.org/10.56884/DWSSCCSE

Abstract

Tracked robots, which have flippers on the front, back, left, and right sides, are expected to be used for disaster investigation because of their high performance to overcome obstacles such as debris and bumps. However, it is difficult for the operator to control the robot because of its high degree of freedom. Therefore, the system that automatically controls the flippers and crawlers is required. In this study, we examined the acquired motion of a tracked robot with flippers to climb a step without using environment information by reinforcement learning. The learned motions are targeted to climb a step efficiently with a small amount of motion and to prevent the large impact when landing on a step. In order to reduce the amount of information required for the decision of the motion, only the information obtained from the internal sensors is used without the information of the surrounding environment. The agent Learned in a simulation environment using multi-body dynamics. First, the robot was trained to climb a step from the front, and the effectiveness of the acquired motion was confirmed from the results of a step climbing simulation and experiments using the trained agent. Then, the motion of the robot for climbing a step from an angle acquired by randomly changing the robot's initial orientation was clarified.

---

Full paper purchase: https://www.istvs.org/proceedings-orders/paper ISTVS members receive three complimentary papers per year: https://www.istvs.org/members