A distributed monitoring approach for human interaction with multi-robot systems

A novel methodology to enable robust and seamless human-robot interaction in the context of multi-robot systems (or swarms) is introduced based on a distributed multi-agent monitoring approach. Through real-time monitoring by each agent of other agents in its observable neighborhood, anomalies (due to malfunctions, cyber-attacks, etc.) in behavior of agents are detected within a probabilistic framework. In the proposed approach, anomaly likelihood estimation is based on how rational/explainable an observed agent's behavior is within the context of the estimated overall situational awareness. A distributed architecture is utilized wherein each agent bases its estimation of other agents' anomaly likelihoods on information currently available to the agent (e.g., from sensors, communications from other agents, etc.).