Center of Mass State Estimation for a 2 DOF Mechanism Based on Sensor Data Fusion

To implement a balance controller on a humanoid robot, state of center of mass (CoM) of the humanoid robot is essential. However due to some features such as joint clearances, link deflections, sensor noises, etc., there is considerable error on determination of CoM. In this regard, this paper investigates the effect of state estimation for a 2 degree of freedom (DoF) mechanism which is a simplified model of the leg of a humanoid robot. Force-torque sensor data and inverted pendulum model are used to predict the position and velocity of the center of mass in the predict step. The estimator also uses the IMU and encoders data in the update step. To study the functionality of the estimator in simulation, white noise is added to the encoder data. The results show that the effect of the white noise in the estimation of position of CoM is reduced. The main objective is to repeat the state estimation method (which was originally implemented on a humanoid robot [1]) on a simple 2 dimensional mechanism to analyze and improve the estimator.