A Sensor Fusion Algorithm: Improving State Estimation Accuracy for a Quadruped Robot Dog

This paper presents a fusion scheme to estimate the state of the quadruped robot dog using the pose estimation of the leg odometer and ORB-SLAM3 algorithm, which is continuous research to provide solutions to the existing problems of internal sensor-based pose state estimation. The problems are described as 1) electromagnetic interference and inaccurate zero position of the motor leading to the accumulation of integral errors in the IMU, and 2) low efficiency and instability of the compensation solutions for the IMU's yaw angular velocity. Aiming at the above problems, the advantages and disadvantages of pose estimation schemes of binocular cameras based on different algorithms are compared and analyzed through data sets experiments and real environment experiments. The Error-State Kalman Filter (ESKF) based fusion framework and formulas are proposed. The comparison fusion experiments using internal and external sensors are conducted with angular velocity compensation and without. The experimental results show a significant improvement in the accuracy and robustness of the pose estimation system, which is and the endpoint error accuracy of the fusion scheme without angular velocity compensation is improved by about 73.5 %.