Security and Safety-Critical Learning-Based Collaborative Control for Multiagent Systems.

Area of Science:

• Control theory
• Artificial intelligence
• Robotics

Background:

• Multiagent systems (MASs) face challenges in communication security and formation safety due to denial-of-service (DoS) attacks, model uncertainties, and environmental barriers.
• Existing control frameworks often struggle to address these combined threats in nonlinear systems.

Purpose of the Study:

• To develop a novel learning-based collaborative control framework for nonlinear MASs.
• To ensure communication security and formation safety under DoS attacks, model uncertainties, and environmental constraints.
• To enhance the resilience and robustness of MASs in dynamic environments.

Main Methods:

• A distributed and decoupled framework integrating cyber-layer and physical-layer designs.
• A resilient control Lyapunov function-quadratic programming (RCLF-QP)-based observer for secure state estimation against DoS attacks.
• Deep reinforcement learning (RL) and control barrier function (CBF) for a safety-critical formation controller.

Main Results:

• The proposed framework successfully ensures secure reference state estimation under DoS attacks.
• A safety-critical formation controller was designed, enabling safe collaborations among uncertain agents.
• Experimental results with autonomous vehicles demonstrated significant improvements in system resilience and robustness.

Conclusions:

• The novel learning-based framework effectively addresses communication security and formation safety in MASs.
• The decoupled cyber-physical design enhances adaptability to DoS attacks and environmental uncertainties.
• The framework shows promise for applications requiring robust and secure multiagent coordination, such as autonomous vehicle formations.