Autonomous Gait Learning of Hexapod Robot Based on Reward Shaping

To improve the autonomous learning ability of robots, a novel autonomous gait learning algorithm based on reward shaping was proposed. First, the mechanical structure of gait synchronous hexapod robot was introduced, with unique multi-layer steel leg design and special mechanical synchronous transmission scheme, which laid a good foundation for robots' gait learning research. Second, a robot gait learning algorithm based on reward shaping was presented, allowing the robot to learn skills in an unknown environment. A concise reward and punishment function, which was based on the principle of simplicity, was designed. Finally, experiments performed on MATLAB and ADAMS joint simulation platform verified the proposed algorithm. Two reasonable convergent results were presented when speed and height, respectively, were the main evaluating indicators. Moreover, the hexapod robot independently learned the difficult skills to move over the blocks with the proposed algorithm.