深層強化学習による歩道環境を移動する四脚ロボットの歩行制御システム

抄録

We developed a gait control system for a quadruped robot using policies trained by deep reinforcement learning. The system switches between multiple policies depending on the situation and handles the problems of fall recovery, staircase adaptation, and power efficiency improvement. We evaluated the gait control system and a navigation system using it through simulation and real-world experiments.