Vision-based Dynamic Gait Stair Climbing Algorithm for Quadruped Robot

Abstract

In order to better serve humans and improve the adaptability of quadruped robots in complex environments, a stable adaptive stair climbing algorithm based on vision is proposed. The terrain and geometry of the stairs can be captured based on the depth camera. The Capture Point (CP) control algorithm based on the Linear Inverted Pendulum Model (LIPM) realizes the centroid stability on the stairs and the reachable domain of the landing point and the desired landing point are found in the camera vision. Finally, a special leg structure is adopted to reduce the phenomenon of sliding and collision, and the posture adjustment algorithm on the stairs is used to liberate the movement space of the front legs. The verification results of the experimental platform show that the quadruped robot can climb 8 steps stably and quickly under the guidance of visual information.