Reinforcement-Learning-Based Robust Controller Design for Continuous-Time Uncertain Nonlinear Systems Subject to...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Designing controllers for uncertain nonlinear systems with control constraints is complex.
• Accurate identification of system uncertainties and input limitations necessitate advanced control strategies.

Purpose of the Study:

• To develop a novel reinforcement learning (RL)-based robust adaptive control algorithm.
• To address continuous-time uncertain nonlinear systems with input constraints.

Main Methods:

• The robust control problem is reformulated as a constrained optimal control problem.
• A single critic neural network (NN) is used to derive the approximate optimal control, differing from typical dual-network approaches.
• Lyapunov's direct method is employed to prove system stability and bounded performance.

Main Results:

• The closed-loop optimal control system and critic NN weights are proven to be uniformly ultimately bounded.
• The derived approximate optimal control guarantees uniform ultimate boundedness for the uncertain nonlinear system.
• Simulation examples demonstrate the algorithm's effectiveness and applicability.

Conclusions:

• The proposed RL-based robust adaptive control algorithm effectively stabilizes uncertain nonlinear systems under input constraints.
• The method offers a robust and practical solution without requiring initial stabilizing control.
• The approach is validated through simulations, showcasing its potential for real-world applications.