Resilient Cooperative Control of Cyber Physical Systems

The advance of embedded systems and networking technology has facilitated a paradigm shift in engineering system design, from centralized to distributed. This shift has lead to significant interest in the design and analysis of multi-agent networks. A multi-agent network, or networked multi-agent system, consists of a set of agents, or nodes that may represent processors, robots, and so on. The agents are equipped with sensing and/or communicating, along with computational resources and possibly actuation. Through a network, the agents share information in order to achieve specific group objectives. Examples of group objectives include consensus, synchronization, formation control, and cooperative load transport. In order for the group objectives to be achieved, distributed algorithms are used to coordinate the behavior of the agents. One of the fundamental challenges in the design of networked multi-agent systems is that the coordination algorithms use only local information, i.e., information obtained by the individual agent through sensor measurements, calculations, or communication with neighbors in the network. Another challenge lies in the fact that not only is each agent a dynamical system, but the network itself is dynamic. Therefore, the distributed algorithms must be designed to handle time-varying network topologies. Information may not be able to be relayed across the dynamic network in a reliable manner. A third challenge is caused by uncertainties introduced by the network and in the implementation of the control and coordination algorithms. More importantly, multi-agent networks, like all large-scale distributed systems, have many entry points for malicious attacks or intrusions. If one or more of the agents are compromised in a security breach, it is crucial for the networked system to continue operating with minimal degradation in performance and the success of the global objective should be assured. To achieve this, it is necessary for the cooperative algorithms to be designed in such a way that they can withstand the compromise of a subset of the nodes and still guarantee some notion of correct behavior at a minimum level of performance. We refer to such a multi-agent network as being resilient to adversaries. This summary describes a framework for resilient cooperative control in networked multi-agent systems using low complexity distributed algorithms for solving consensus and synchronization problems in dynamical systems. The framework combines methods from distributed computing and control engineering to devise coordination algorithms that ensure that group objectives such as consensus and synchronization are satisfied even in the presence of adversaries. The work can contribute to providing scalable computational methods for resilient cooperative control in the presence of adversaries while ensuring robustness to real-world system behavior and interactions in dynamic networks.