Discontinuous Rod Spanning Motion Control for Snake Robot based on Reinforcement Learning

The high redundant degree of freedom characteristics and limited moments of the snake robot pose difficulties for the discontinuous rod spanning motion, and it is difficult to build an accurate control model using traditional control methods. In this work, a reinforcement learning-based method is proposed to design a controller for the discontinuous rod spanning motion of the snake robot. Specifically, we present an RL-based controller that is divided into two tandem phases for swing and grab bar control and is trained using a proximal policy optimization (PPO) algorithm. Experimental results show that the proposed RL-based controller swings to the target height faster compared to the energy-pumping-only method and achieves a swinging grab bar.