Study on the Rolling Motion of a Snake-Like Robot That Transforms into a Parallel Two-Wheeled Vehicle Using Deep Reinforcement Learning

Abstract

Snake-like robots have higher mobility than wheeled robots. However, there is an issue of increased power consumption when they reproduce the undulating locomotion like biological snakes because of a lot of servomotors driving on joints. We propose an efficient method that integrates the center-of-gravity (COG) shifting for the navigation of the robot to address the aforementioned problem. In the proposed method, a snake-like robot is transformed into the shape of a tire to realize the parallel two-wheeled vehicle, and the COG is changed by deforming the links to generate a rolling motion. This implementation allows for the choice between rolling and undulating locomotion depending on whether the level or rough ground. Previous research has used gravity data from an acceleration sensor as feedback, but this has led to problems with maintaining straight-line stability when road conditions change. This paper presents a controller design method using deep reinforcement learning (RL) to achieve robust traveling by the rolling motion.