Game-Based Backstepping Design for Strict-Feedback Nonlinear Multi-Agent Systems Based on Reinforcement Learning.

Area of Science:

• Control Theory
• Artificial Intelligence
• Robotics

Background:

• High-order nonlinear multi-agent systems present significant control challenges due to unknown dynamics and input saturation.
• Robust control strategies are essential for ensuring stability and performance in complex systems.

Purpose of the Study:

• To propose a game-based backstepping control method for high-order nonlinear multi-agent systems.
• To achieve robust control by addressing unknown dynamics and input saturation using reinforcement learning.

Main Methods:

• Employed reinforcement learning (RL) to find the saddle point solution of the tracking game.
• Utilized policy iteration and a single network adaptive critic (SNAC) architecture to solve the Hamilton-Jacobi-Isaacs (HJI) equation.
• Approximated unknown nonlinear dynamics using state differences from a command filter.

Main Results:

• Successfully regulated the final equilibrium point by adjusting agent proportions in the game.
• Established a sufficient condition guaranteeing uniform ultimate boundedness for the entire system and its subsystems.
• Simulation results validated the effectiveness of the proposed control method.

Conclusions:

• The proposed game-based backstepping control method effectively handles high-order nonlinear multi-agent systems with uncertainties.
• Reinforcement learning provides a robust approach to achieve stable control under input saturation.
• The method demonstrates practical applicability and ensures system stability.