Moving Horizon Body State Estimation for Quadruped Robot with Multi-Sensor Information Fusion

The position and velocity information of the quadruped robot are essential for controlling. With accurate estimation, more precise feedback can be provided, thus the performance of the robot will be improved. Single inertial navigation solution (INS) or kinematic analysis cannot meet the requirement in long-term motion because of the presence of drift in sensoring or noise resulting from impact and oscillation. Meanwhile, in practical cases, states of the robot are often constrained by conditions. But traditional estimation methods, such as Kalman filtering, cannot deal with constraints, resulting that the estimation may violate the physical reality. As an optimal estimation method with superior performance, moving horizon estimation (MHE) carry out online estimation by solving the optimal problem with constraints explicitly. It has unique advantages in processing nonlinear system, uncertain measurement system and state constrained system. This paper proposes an estimation method for multi-sensor fusion. The moving horizon estimation method integrates inertial navigation and robot kinematics, overcoming single channel solution’s shortcomings, and makes the optimal estimation for the legged robot. The results of simulation prove the effectiveness of the method.