A Novel Foot Contact Probability Estimator for Biped Robot State Estimation

State estimation is an important part of biped robot control, but unreliable foot contact estimation would lead to inaccurate state estimation result. In this paper, to reduce the state estimation error caused by the inaccurate contact estimation, we propose a novel simplified contact probability estimator based on the force/torque sensor mounted on the foot. The contact probability is used to tune the covariance matrices of the extended Kalman filter, and the parameters of the probability estimator are optimized iteratively through minimizing the error between state estimation result and ground truth measurement. The experimental result on BHR-6P biped robot shows that the proposed method can effectively reduce the state estimation error.