Modeling and Control Design for a Musculoskeletal Robot via Adaptive Dynamic Programming

In this study, for the musculoskeletal robot system, we aim at optimizing the angle tracking control based on the system model. First, we analyze the driving principle of muscles and establish a bionic muscle dynamics model. Further, based on the geometric relationship between muscles and skeletons, we establish the kinematics model and dynamics model of the musculoskeletal robot system. The adaptive dynamic programming (ADP) algorithm is used for solving the Hamilton-Jacobi-Bellman (HJB) equation. Based on the proposed updating law, the neural network is trained to approximate the solution. Using the Lyapunov’s direct method, we can prove that the system is stable with the proposed controller. Furthermore, simulation indicates that this presented control law is feasible. Note to Practitioners—The motivation of this study is to investigate the dynamic modeling and optimal control design of musculoskeletal robots. Existing research on musculoskeletal robots has mostly focused on the mechanical design, with less consideration given to the analysis of system dynamics. Typically, model-free control algorithms are employed, which affect the precise control performance of the system. This paper aims to develop a mathematical representation of musculoskeletal robots based on the biological and mechanical characteristics of muscles and bones. The proposed method can optimize the tracking performance of the system, avoiding repetitive trial and error. Aseries of simulation experiments have been conducted to validate the effectiveness and applicability of this method.