Reinforcement Learning-Based Cooperative Control for Nonlinear Multiagent System With State and Control Input...

Area of Science:

• Control Systems Engineering
• Robotics
• Artificial Intelligence

Background:

• Multiagent systems require robust cooperative tracking control.
• Challenges include unknown dynamics, state, and input constraints.

Purpose of the Study:

• To develop a novel, model-free control framework for cooperative tracking in multiagent systems.
• To address unknown dynamics, state constraints, control input constraints, optimal performance, and convergence rate constraints.

Main Methods:

• Utilized the mean value theorem to handle control input constraints.
• Employed a performance function with a barrier Lyapunov function for transient performance and state constraints.
• Leveraged actor-critic neural networks for near-optimal solutions with unknown dynamics.

Main Results:

• The proposed control scheme is completely model-free.
• Rigorous Lyapunov stability analyses confirmed satisfaction of all constraints.
• Cooperative tracking errors were minimized in finite time with a desired decay rate.

Conclusions:

• The novel framework effectively manages complex constraints in multiagent systems.
• Simulation and hardware tests validated the proposed control strategy's effectiveness.
• Achieved guaranteed transient performance, finite-time convergence, and optimal tracking.