Analysis of Position and State Estimation of Quadruped Robot Dog Based on Invariant Extended Kalman Filter

摘要

Abstract: Compared with the state estimation of quadruped robots based on external sensors such as camera and lidar, the state estimation based on body sensors can provide high-frequency and stable odometer estimation. By analyzing the state estimation methods of the legged robot based on the body sensor, the invariant extended Kalman filter (IEKF) based on the body sensor is determined to conduct the state estimation analysis of the quadruped robot. Through various path tracking experiments in simulation and real environment, the influence of travel speed, travel distance and different steering angles on the position state estimation results was analyzed, and the IEKF model was optimized by compensating the angular velocity. Experiments show that within the set speed range, after adding angular velocity compensation, the position estimation accuracy error of the robot dog is well controlled and is less than 1%.