New Approach to Center of Mass Estimation for Humanoid Robots Based on Sensor Measurements and General LIPM

We present a center of mass estimator based on 6-axis force torque sensors and encoders. Acceleration of Center of Mass (CoM) is calculated using the total external forces. It is integrated to predict the CoM position and velocity. The forward kinematic equation and general Inverted Pendulum Model is used without any constraints on CoM height to update the states. The general model and equations handle changes of the height of CoM. This estimator is implemented on a simplified model of a humanoid robot in simulation. It is assumed that CoM of each link has some unknown offset errors. The estimator successfully compensates the offset errors besides the white noise in encoders and force torque sensors data, and finally estimates the CoM position on the ground.

I. Introduction

The position of the center of mass (CoM) plays an important role in the model-based control of humanoid robots e.g. in model predictive controllers (MPC) [1]. Also Capture Point (CP) can be approximately calculated by the CoM position and velocity [2]. In addition, controlling humanoid robots with CP needs the CoM position to be known precisely [3], [4] and [5].