Resilient Autonomous Control of Distributed Multi-agent Systems in Contested Environments

摘要

An autonomous and resilient controller is proposed for leader-followermulti-agent systems under uncertainties and cyber-physical attacks. The leaderis assumed non-autonomous with a nonzero control input which allows changingthe team behavior or mission in response to environmental changes. A two-layerresilient learning-based control protocol is presented to find optimalsolutions to the synchronization problem in the presence of attacks and systemdynamic uncertainties. In the first security layer, an observer-baseddistributed H-infinity controller is designed to prevent propagating theeffects of attacks on sensors and actuators throughout the network, as well asattenuate the effect of these attacks on the compromised agent itself.Non-homogeneous game algebraic Riccati equations are derived to solve theH-infinity optimal synchronization problem and an off-policy reinforcementlearning is utilized to learn their solution without requiring any knowledge ofthe agent's dynamics. In the second security layer, a trust-confidence baseddistributed control protocol is proposed to mitigate attacks that hijack theentire node and attacks on communication links. A confidence value is definedfor each agent based on only its local evidence. The proposed RL algorithmemploys the confidence value of each agent to indicate the trustworthiness ofits own information and broadcast it to its neighbors to put weights on thedata they receive from it during and after learning. If the confidence value ofan agent is low, it employs a trust mechanism to identify compromised agentsand remove the data it receives from them from the learning process. Simulationresults are provided to show the effectiveness of the proposed approach.