Value Iteration-Based Adaptive Fuzzy Backstepping Optimal Control of Modular Robot Manipulators via Integral Reinforcement Learning

An adaptive fuzzy backstepping optimal control method is developed for modular robot manipulators (MRMs) via value iteration (VI). This paper adopts joint torque feedback (JTF) technique to construct subsystem dynamics model, and the state space description is deduced. According to fusion function which contains the error in joint angular velocity and position, the cost function is established. The integral reinforcement learning (IRL) is integrated into the VI algorithm, which solves the optimal tracking control issue without system drift dynamics. For purpose of improving the control effect, the optimal tracking control issue of manipulator can be reconsidered as the optimal compensation issue which adopting the local dynamics information. Then the uncertainty in the model can be compensated by an adaptive fuzzy backstepping compensation controller which is constructed by fuzzy logic system (FLS) and backstepping control method. The optimal compensation control strategy is adopted to deal with the interconnected dynamic coupling (IDC), which contains global information about each joint. Based on the VI algorithm and adaptive dynamic programming (ADP) method, an effective solution of Hamiltonian-Jacobi-Bellman (HJB) equation is presented. According to Lyapunov theorem, the trajectory tracking error is uniformly ultimately bounded (UUB) by using the adaptive fuzzy backstepping optimal control method. Finally, the effectiveness of the proposed method is verified by experiments.