Underwater gait planning and control of HUHR for wall climbing and transition based on deep reinforcement learning

In this paper, we present a gait planning and control method for a hybrid-driven underwater hexapod robot (HUHR) to realize continuous legged locomotion between different walls based on deep reinforcement learning (DRL), while tracking the desired forward speed and yaw angle. The method allows the robot to learn the mapping relationship between the rotation angle of each leg and the states, so that our robot can generate and adjust gait autonomously during the movement. Meanwhile, we realize coordinated control of thrusters and legs, which reduces the action space dimension and training difficulty of hybrid-driven robots. Firstly, we characterize the touchdown state of each C-shape leg through coordinate analysis. Then, we design the details about the DRL framework and analyse how to set appropriate thrust. Finally, we train and test our policies in Gazebo. The test results verify the effectiveness of the proposed method.