Decentralized Intrusion Detection in Cooperative Multi-Agent Systems

We address the problem of detecting faulty behaviors of robots belonging to a multi-agent system. Our objective is to develop a scalable architecture that can be adopted to realize a completely decentralized intrusion detector monitoring the agents' behavior. We want the solution to be independent from the set of ``rules'' describing the interaction among the agents, and from their dynamics; (non-invasive) mainly based on HW/SW components that are already present on-board of each agent. We focus on systems with decentralized cooperation schemes where cooperation is obtained by sharing a set of ``rules'' by which each agent plans its next ``action'' and where some of the agents may act not according to the rules due to spontaneous failure, tampering, or malicious introduction.