Vision-Based Motion Planning Algorithm for Quadruped Robot of Stair Terrain

Quadruped robots can complete action tasks in various complex terrain because of their significant adaptability. Since staircase are common in practical scenarios, achieving stable locomotion on stairs is important for quadruped robots. Conventional motion planners are not as effective on stairs because of their discontinuous structures. However, by using terrain information obtained through perception, perceptual motion planning method performs better on this challenging scenario. Based on perceptual method, this paper presents a stable stair climbing motion planning algorithm for quadruped robots using elevation maps. By leveraging stair structure information derived from elevation maps, we design a method to generate the the center of mass (CoM) motion trajectory based on the virtual zero-moment point (ZMP) stability criterion and optimize foothold selection. Additionally, a combined third-order Bézier curve is introduced to generate swing leg trajectories based on predicted footholds and stair structure. In order to verify the efficiency of the proposed algorithm, an autonomously climbing stairs of varied geometric shapes simulation scenario is constructed, in which a Unitree’s go1 quadruped robot is introduced.