Dead reckoning of a biped robot on various terrain by Kaiman filter adaptive to ground reaction force

A novel Kaiman filter to estimate position of a biped robot is proposed. It combines the kinematics and the double-integral of acceleration, only using internal sensors and achieving high-rate estimation. The kinematics computation is rooted to the anchoring pivot, which is the most invariant point in the foot with respect to the ground. The idea is the same with the authors previous method, but the estimation accuracy has been largely improved by referring to the ground reaction force. Namely, the anchoring pivot is estimated based on both the velocity information and the force information. The efficacy of the proposed method is verified through simulations of walking and hopping motions.