Pose Estimation Algorithm for Quadruped Robots Based on Multi-Sensor Fusion

Aiming to address the problem of a low pose estimation accuracy of quadruped robots in complex terrains and during high-speed movements, this paper proposes a pose estimation algorithm that integrates binocular cameras, joint encoders, and inertial measurement unit (IMU). The proposed algorithm combines a forward kinematic model of a quadruped robot with the IMU trajectory data as a sub-filter for joint filtering. In addition, the forward kinematic model provides precise foot-end position information on a robot, and the IMU contributes the global posture, velocity, and position data. Further, the binocular camera data are fused with the IMU posture data, forming another sub-filter for joint filtering to enhance environmental perception. Finally, the results obtained from the two sub-filters are integrated to achieve a globally optimal estimation, providing the robot’s pose information. The proposed algorithm’s accuracy and robustness are verified by experiments on a self-developed robotic platform and a comparative analysis with the VIO data.