Non-linear ZMP based state estimation for humanoid robot locomotion

摘要

This article presents a novel state estimation scheme for humanoid robot locomotion using an Extended Kalman Filter (EKF) for fusing encoder, inertial and Foot Sensitive Resistor (FSR) measurements. The filter's model is based on the non-linear Zero Moment Point (ZMP) dynamics and thus, coupling the dynamic behavior in the frontal and the lateral plane. Furthermore, it provides state estimates for variables that are commonly used by walking pattern generators and posture balance controllers, such as the Center of Mass (CoM) and the linear time-varying Divergent Component of Motion (DCM) position and velocity, in the 3-D space. Modeling errors are taken into account as external forces acting on the robot in the acceleration level. In addition, an observability analysis for the non-linear system dynamics and the linearized discrete-time EKF dynamics is presented. Subsequently, by utilizing ground-truth data obtained from a vicon motion capture system with a NAO humanoid robot, we demonstrate the effectiveness and robustness of the proposed scheme contrasted to the linear filters, even in the case where disturbances are introduced to the system. Finally, the proposed approach is implemented and employed for feedback to a real-time posture controller, rendering a NAO robot able to walk on an outdoors inclined pavement.