Distributed state estimation for interdependent and switching multi-agent systems

This paper addresses the problem of multi-agent distributed state estimation in switching networks over directed graphs. Specifically, we consider a novel estimation setting for a linear continuous-time system that is broken down into subsystems, each of which is locally estimated by the corresponding agent. This task is undertaken despite the complexities due to interdependencies on both cyber and physical levels, and due to the fact that the system is switching, i.e., the different subsystems/agents can activate (e.g., to accomplish some specific task) or deactivate (e.g., due to a fault) during a transient that ends with a cutoff time, unknown to the agents, after which the topology becomes fixed. In particular, by exploiting the negativizability property – the pair $(A,C)$ is negativizable if there is a feedback gain $K$ such that $A-KC$ is negative definite – each agent is able to locally perform the calculation of its own estimation gain matrix. The paper is complemented by the convergence analysis of the estimation error executed by leveraging on nonsmooth analysis and simulations to prove the effectiveness of the proposed results.