Optimal Tracking Control of a Nonlinear Multiagent System Using Q-Learning via Event-Triggered Reinforcement...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Robotics

Background:

• Multi-agent systems (MASs) often face challenges in tracking control due to unknown nonlinear dynamics.
• Traditional control methods can be computationally intensive and require extensive system knowledge.
• Event-triggered control strategies offer potential for reduced communication and computation.

Purpose of the Study:

• To develop an optimal control tracking method for unknown nonlinear MASs.
• To reduce computational load and communication frequency using an event-triggered technique.
• To enhance tracking performance through an internal reinforcement Q-learning algorithm.

Main Methods:

• An event-triggered technique is combined with the internal reinforcement Q-learning (IrQL) algorithm.
• A Q-learning function is calculated using an internal reinforcement reward (IRR) formula.
• A neutral reinforce-critic-actor (RCA) network is designed for performance assessment and online learning.
• An event-triggered weight tuning rule modifies actor neural network parameters based on triggering conditions.

Main Results:

• The proposed event-triggered IrQL method effectively addresses tracking control in unknown nonlinear MASs.
• The event-triggered approach significantly reduces data transmission rates and computational burden compared to time-triggered methods.
• Lyapunov-based convergence analysis confirms the stability and performance of the reinforce-critic-actor neural network.
• Demonstrated efficiency and accessibility of the data-driven approach through an illustrative example.

Conclusions:

• The developed event-triggered optimal control strategy offers an efficient and data-driven solution for tracking control in MASs.
• This method minimizes computational and communication overheads, making it suitable for resource-constrained applications.
• The internal reinforcement Q-learning and RCA network provide robust online learning and performance assessment capabilities.