Nonzero-sum game-based online off-policy approximate optimal control of modular robotic manipulator system via data-driven method

This paper proposes a data-driven online off-policy approximate optimal control method within a nonzero-sum game framework for trajectory tracking tasks in modular robotic manipulator systems. To address the problem of partially inaccessible system dynamics, a data-driven neural network identifier is constructed to estimate subsystem models in modular robotic manipulator systems. In environments such as aerospace applications, modular robotic manipulator systems require the realization of optimal control strategies to reduce energy consumption while accomplishing trajectory tracking tasks. However, traditional approximate dynamic programming methods often converge to local optima. Therefore, to solve the optimal control problem of the distributed multi-controller system of modular robotic manipulator systems, an online off-policy approximate dynamic programming method is integrated into a nonzero-sum game framework to minimize the cost of the entire system. Finally, system stability is proven via Lyapunov theory, and experimental validation further confirms the effectiveness of the proposed method.