Adaptive Locomotion Control of Hexapod Robot Based on Deep Reinforcement Learning and Proprioception

The hexapod robot is widely used for outdoor missions due to its superior traversability. At present, hexapod robots use external sensors to obtain environmental information, but external sensors are susceptible to natural factors such as illumination, which will lead to sensor failure in serious cases. This paper investigates the normal movement of hexapod robot in unstructured environment solely relying on internal sensors. The information obtained from internal sensors meets minimum requirements for normal locomotion of hexapod robot, just like walking with closed eyes for human beings. This method uses deep reinforcement learning to train the locomotion strategy of hexapod robot in simulation environment, which enables hexapod robot to walk steadily at a fixed speed in an unstructured environment and has certain obstacle avoidance ability.