Reinforcement Learning-Based Optimal Stabilization for Unknown Nonlinear Systems Subject to Inputs With Uncertain...

Area of Science:

• Control Theory
• Artificial Intelligence
• Nonlinear Systems

Background:

• Stabilizing unknown nonlinear systems with input constraints is challenging.
• Existing methods may struggle with uncertainty and saturation nonlinearities.

Purpose of the Study:

• To develop a novel reinforcement learning strategy for optimal stabilization.
• To address unknown nonlinear systems with uncertain input constraints.

Main Methods:

• A two-part control algorithm: online learning optimal control and neural network (NN) compensation.
• Utilizing a Luenberger observer with recurrent NN to approximate system dynamics.
• Solving the Hamilton-Jacobi-Bellman equation via a critic NN for nominal systems.
• Employing a feedforward NN compensator for uncertain input constraints (saturation nonlinearities).

Main Results:

• Guaranteed uniform ultimate boundedness of the closed-loop system via Lyapunov stability analysis.
• Effective compensation for uncertain input constraints.
• Demonstrated stabilization scheme effectiveness through simulation studies.

Conclusions:

• The proposed reinforcement learning strategy effectively stabilizes unknown nonlinear systems with uncertain input constraints.
• The integration of online learning and NN compensation provides a robust solution.
• The method offers a promising approach for complex control problems.