MOTION PLANNING AND CONTROL FOR WHEELED-LEGGED QUADRUPEDAL ROBOT LOCOMOTION

Abstract

This thesis aims to develop an effective motion planning and control framework for the locomotion of a wheeled-legged quadruped robot. The pure-legged locomotion framework is first presented in this thesis, followed by the framework for the hybrid locomotion. First, a bio-inspired method is presented for legged locomotion. Based on a simple Central Pattern Generator (CPG) model, this method produces motions capable of realizing multiple gaits and smooth gait transitions. Next, a motion planning and control framework is developed to address the hybrid locomotion problem of wheeled-legged robots. This framework comes as a hierarchical structure with foothold planning, Center of Mass (CoM) trajectory optimization, and whole-body control. Finally, to predict the gait patterns in hybrid walking-driving mode more intelligently, a gait pattern planner based on deep reinforcement learning (DRL) techniques is proposed.