Cooperative Online Learning for Multiagent System Control via Gaussian Processes With Event-Triggered Mechanism

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Multiagent systems (MASs) with unknown dynamics pose significant control challenges.
• Gaussian process (GP) regression offers flexible nonlinear function modeling and prediction error bounds for uncertainty inference.
• Online learning enhances GP model predictions by incorporating new data during operation.

Purpose of the Study:

• To investigate an online cooperative learning algorithm for multiagent systems (MASs) control.
• To develop an event-triggered data selection mechanism for improved data efficiency in GP-based MAS control.
• To validate the practical convergence and tracking performance of the proposed learning-based control strategy.

Main Methods:

• Online cooperative learning algorithm utilizing Gaussian process (GP) regression.
• Event-triggered data selection mechanism inspired by centralized event-trigger (CET) analysis.
• Lyapunov theory for validating system convergence and tracking performance guarantees.

Main Results:

• The proposed online learning algorithm ensures practical convergence of MASs.
• Guaranteed tracking performance is achieved for the multiagent systems.
• The event-triggered mechanism effectively reduces model update frequency and enhances data efficiency.
• Zeno behavior is demonstrably excluded for individual agents.

Conclusions:

• The developed event-triggered online learning method is effective for cooperative control of MASs with unknown dynamics.
• The approach enhances prediction accuracy and control performance through efficient data utilization.
• Theoretical guarantees for convergence and tracking performance are provided using Lyapunov stability analysis.