FRL-SLAM: A Fast, Robust and Lightweight SLAM System for Quadruped Robot Navigation

Abstract

The quadruped robots have been outstood as potential mobile platforms for logistics transportation and hazardous environment explorations because of their extraordinary locomotion capability. quadruped robots are desirable as they can maneuver through stairs, grotto and grassland effortlessly in unstructured scenes. However, quadruped robots often fail to reach designated destinations due to the SLAM module is disturbed if the trunk of a legged robot will swing strongly during locomotion. Therefore, the current quadruped robot navigation algorithms are not satisfactory. In this paper, we proposed FRL-SLAM, a fast, robust and lightweight visual-inertial SLAM system, to address these issues. The FRL-SLAM can accurately estimate the robot pose with a RealSense D435i camera in the case of violent shaking. We estimate the performance of FRL-SLAM on public datasets and real-world quadruped robot and compare against other state-of-the-art algorithms. Our experiments show that FRL-SLAM is as accurate as the state-of-the-art SLAM system and significantly more fast.