Concurrent Learning Critic-Only NN-Based Robust Approximate Optimal Control of Nonlinear Systems With Experimental Verification

This article addresses the optimal control problem for a class of continuous-time nonlinear systems with time-varying bounded disturbances. A novel approximate optimal control policy that incorporates a continuous robust control command is developed within the framework of adaptive dynamic programming. It makes the controlled system robust to the bounded time-varying disturbance rather than just the disturbance that vanishes as the system state converges. Moreover, an improved concurrent learning neural network (NN) weight updating algorithm that involves a switching mechanism is presented, and it can work without persistent/finite assumptions. Then, with the uniformly ultimately bounded stability guaranteed by Lyapunov theory, the closed-loop controlled system achieves approximate optimality and strong robustness. The superiority and effectiveness of the suggested control approach have been demonstrated via robotic trajectory tracking experiments.