Actor–Critic Learning Based Pid Control for Robotic Manipulators

peer reviewedIn this paper, a reinforcement learning structure is proposed to auto-tune PID gains by solving an optimal tracking control problem for robot manipulators. Taking advantage of the actor-critic framework implemented by neural networks, optimal tracking performance is achieved while unknown system dynamics are estimated. The critic network is used to learn the optimal cost-to-go function while the actor-network converges it and learns the optimal PID gains. Furthermore, Lyapunov’s direct method is utilized to prove the stability of the closed-loop system. By that means, an analytical procedure is delivered for a stable robot manipulator system to systematically adjust PID gains without the ad-hoc and painstaking process. The resultant actor-critic PID-like control exhibits stable adaptive and learning capabilities, while delivered with a simple structure and inexpensive online computational demands. Numerical simulation is performed to illustrate the effectiveness and advantages of the proposed actor-critic neural network PID control