Locomotion Planning for Quadruped Robot Walking on Lunar Rough Terrain

This paper focuses on the locomotion planning for a quadruped robot walking on the lunar rough terrain. Firstly, the detailed terrain data of the explorable area acquired by the navigation camera is filtered. The terrain is afterwards triangular meshed and reconstructed as a simplified triangular grid model with terrain features retained. Then, the reinforcement learning method is used to plan the path of the robot in the grid-based environment. It employs terrain relief and roughness as the rewards, therefore intelligently determining the optimal detection route with maximum cumulative reward. Finally, gait planning is carried out to make the legs actuate adaptively to the path. Particularly, the step sequence is adjusted with different steering angles, and the footsteps are decided based on the robot mechanism constraints and uneven terrain conditions. Numerical simulations illustrate the walking process of the quadruped robot. The results show that the robot can learn the optimal path with fewer trunk undulations, and generate continuous, stable, and safe gaits. It proves that the locomotion planning method can effectively improve the mobile stability, efficiency, and adaptability of the quadruped robot when walking on the lunar surface.