Integral Reinforcement Learning-Based Dynamic Event-Triggered Nonzero-Sum Games of USVs

Area of Science:

• Robotics and Control Systems
• Game Theory
• Artificial Intelligence

Background:

• Unmanned Surface Vehicles (USVs) require robust control strategies for navigation and task execution.
• Achieving Nash equilibrium in Nonzero-Sum (NZS) games is crucial for multi-agent coordination.
• State and input constraints pose significant challenges in USV control systems.

Purpose of the Study:

• To develop an Integral Reinforcement Learning (IRL) method for dynamic event-triggered NZS games.
• To enable USVs to achieve Nash equilibrium while adhering to state and input constraints.
• To enhance computational efficiency and reduce network load through event-triggered control.

Main Methods:

• A mapping function was designed to ensure USV states and controls remain within a safe operational environment.
• IRL-based coupled Hamilton-Jacobi equations were derived, bypassing the need for explicit system dynamics.
• A dynamic event-triggered control mechanism was developed, improving upon static approaches.
• Critic neural networks were employed to approximate value functions and control policies for each agent.

Main Results:

• The proposed IRL method successfully achieves Nash equilibrium for USVs under constraints.
• The dynamic event-triggered approach conserves computational resources and network bandwidth.
• Rigorous proofs confirmed the uniform ultimate boundedness of system states and weight estimation errors.
• Simulation experiments validated the effectiveness of the developed control strategy.

Conclusions:

• The integral reinforcement learning method provides an effective solution for constrained NZS games in USV systems.
• Event-triggered control significantly optimizes resource utilization in cooperative USV operations.
• The approach demonstrates strong theoretical guarantees and practical applicability through simulations.