Multi-terrain Motion Control Method for Quadruped Robot Based on Reinforcement Learning

Quadruped robot has become a research hotspot because of its good flexibility and adaptability. Most of the current research work has been conducted on flat ground or simple slopes, and it is difficult to achieve robust locomotion in complex environment and multi-terrain. In this paper, a motion control method combining Central Pattern Generator (CPG) and reinforcement learning is proposed. CPG is used to simulate the leg rhythm information of quadruped animals and generate reference motion of robots. By building a multilayer perceptron (MLP) to realize the state detection of the robot, and using reinforcement learning algorithms to optimize the parameters of the network, the robot’s attitude adaptive adjustment and motion balance control can be realized. The simulation experiment verifies that the robot using this controller can move smoothly on flat ground, mountainous terrain with strong self-adaptive capability.