Reinforcement Learning-Based Control Strategy for Multi-Agent Systems Subjected to Actuator Cyberattacks During Affine Formation Maneuvers

In this research, we investigate the reinforcement learning-based control strategy for second-order continuous-time multi-agent systems (MASs) subjected to actuator cyberattacks during affine formation maneuvers. In this case, a long-term performance index is created to track the MASs tracking faults using a leader-follower structure. In order to approximate the ideal solution, which is challenging to find for systems vulnerable to cyberattacks during time-varying maneuvers, a critical neural network is used. The distributed control protocol is obtained, and the long-term performance index is minimized, using an actor neural network strengthened with critic signals. The actor-critic neural networks calculate unknown dynamics and the severity of attacks on the MAS actuators. The Nussbaum functions are applied to address this issue since attacks can result in a loss of control direction. The stability of the closed-loop system has been emphasized with the use of a Lyapunov candidate function. The performance of the suggested strategy is then supported by a numerical simulation.

Automated summary

In this research, a reinforcement learning-based control strategy is investigated for second-order continuous-time multi-agent systems (MASs) under actuator cyberattacks during affine formation maneuvers. To approximate the ideal solution for systems vulnerable to cyberattacks during time-varying maneuvers, a critical neural network is utilized. The actor-critic neural networks are employed to calculate unknown dynamics and severity of attacks on the MAS actuators. The stability of the closed-loop system is emphasized using a Lyapunov candidate function. The suggested strategy is supported by a numerical simulation.