Reinforcement Learning-Based Decentralized Safety Control for Constrained Interconnected Nonlinear Safety-Critical...

Area of Science:

• Control Theory
• Artificial Intelligence
• Nonlinear Systems

Background:

• Decentralized safety control (DSC) is crucial for interconnected nonlinear safety-critical systems.
• Existing methods often struggle with asymmetric input and security constraints.
• Reinforcement learning (RL) offers a promising avenue for adaptive control strategies.

Purpose of the Study:

• To develop a novel DSC strategy for constrained interconnected nonlinear safety-critical systems.
• To address challenges posed by asymmetric input and security constraints.
• To ensure stable system operation and optimal policy learning within safe domains.

Main Methods:

• Constructed improved performance functions for actuator estimates in auxiliary subsystems.
• Transformed the decentralized control problem with security and asymmetric input constraints into an equivalent problem with only asymmetric input constraints using barrier functions.
• Applied Lyapunov theory to analyze system stability.

Main Results:

• The proposed method ensures that safety-critical systems operate and learn optimal DSC policies within their safe global domains.
• The optimal control strategy guarantees that the entire system is uniformly ultimately bounded (UUB).
• All signals in the closed-loop auxiliary subsystem are demonstrated to be uniformly ultimately bounded.

Conclusions:

• The developed decentralized safety control strategy effectively handles asymmetric input and security constraints in nonlinear systems.
• The reinforcement learning-based approach ensures system stability and optimal control policy convergence.
• Simulation results validate the practical effectiveness of the proposed method for safety-critical applications.