Adaptive Control for a 3D Snake-Like Robot Based on Mutual Supervised Reinforcement Learning

摘要

Snake-like robots are suitable for complex scenarios due to their redundancy and flexibility. Their locomotion depends on switching between different gaits. The working scenarios of snake-like robot are usually energy-constrained. Traditional methods usually set fixed gaits, but these approaches cannot adapt to different scenarios and switch gait to improve energy efficiency and reach the target. In this paper, we propose an adaptive gait control method for snake-like robots based on mutually supervised reinforcement learning, and test our method on Coppeliasim simulator and a real robot. Tests have demonstrated that our approach increases the snake-like robot’s target arrival rate by at least 37.93% compared to three other approaches in the literature, and reduces energy consumption by 10.6%.