Joint proprioception acquisition strategy based on joints-muscles topological maps for musculoskeletal humanoids

Many humanoids have been developed, but more complicated and flexible humanoids must be developed, in order to realize more natural and various motions like humans. However, it is difficult to measure directly joint posture in the multi-dofs joint of such robots (e.g. a hip spherical joint) because of its complicated structure. This paper describes an estimation method for tendon-driven joint postures of these complicated multi-dofs joints, only based on information of muscles' length relative displacement data during joints movement. We regard this posture estimation problem as a pattern matching problem in the mapping space from joint posture to muscle lengths and solve this problem by using very simple searching algorithms. Furthermore, this paper describes a strategy for deciding muscles motor command to acquire its joint proprioception by the proposed joint posture estimation method only based on information of muscles' length relative displacement data. Finally, we confirmed the feasibility of the proposed estimation method by applying this algorithm to the real tendon-driven musculoskeletal humanoid Kojiro.