Iterative learning control for a musculoskeletal arm: Utilizing multiple space variables to improve the robustness

In this paper, a new iterative learning control method which uses multiple space variables for a musculoskeletal-like arm system is proposed to improve the robustness against noises being included in sensory information. In our previous works, the iterative learning control method for the redundant musculoskeletal arm to acquire a desired endpoint trajectory simultaneous with an adequate internal force was proposed. The controller was designed using only muscle space variables, such as a muscle length and contractile velocity. It is known that the movement of the musculoskeletal system can be expressed in a hierarchical three-layered space which is composed of the muscle space, the joint space and the task space. Thus, the new iterative learning control input is composed of multiple space variables to improve its performance and robustness. Numerical simulations are conducted and their result is evaluated from the viewpoint of the robustness to noises of sensory information. An experiment is performed using a prototype of musculoskeletal-like manipulator, and the practical usefulness of the proposed method is demonstrated through the result.