A joint-space controller based on redundant muscle tension for multiple DOF joints in musculoskeletal humanoids

Abstract

To achieve contact tasks with musculoskeletal humanoids, adaptive motion by muscle tension control and robustness against actuator malfunction is important. In this paper, we develop a tension-based joint-space controller for musculoskeletal multiple DOF joints. Joint angle estimation is integrated with the controller, enabling application to spherical joints and spine structure whose joint angle cannot be directly measured. Furthermore, by utilizing the muscle redundancy, a fault tolerant controller is enabled. For evaluation we develop the head and neck of the musculoskeletal humanoid “Kengoro”. We demonstrate by motion generating experiments that the controller is valid and that joint torque estimation is improved compared with a previous controller based on muscle length. Toward an application for contact tasks, we show that contact detection on unknown environments is achieved utilizing the estimated joint torque.