보행 주파수 변경이 가능한 6족 보행 로봇의 학습 기반 이동 제어

This paper proposes an end-to-end walking control methodology for a hexapod robot based on reinforcement learning, which is bio-inspired by the walking characteristics of ants, to enhance stability in highly uncertain terrains. Specifically, the study aims to improve walking stability by dynamically adjusting the gait frequency according to walking speed and terrain conditions. First, the study selects the appropriate body shape and leg structure and arrangement to enhance the mobility stability and efficiency of the hexapod robot. Next, a gait frequency determination mechanism based on the walking frequency adjustment mechanism of ants is established, and a reinforcement learning strategy is designed to enable walking using proprioceptive information. Finally, the policy is trained in Isaac Gym, and the walking stability of the trained policy is comparatively analyzed and verified through a Gazebo Sim-to-Sim process.