An Initial Assessment of Plan-Recognition-Based Coordination for Multi-Agent Teams

Plan recognition remains a largely unexplored paradigm for facilitating coordination. In this paper, we begin to explore domain, task, and agent characteristics which impact upon the utility of using plan recognition for coordinating multiple agents and, in particular, collections of agents organized into competing teams. Agents in our research are supplied plan-recognition capabilities in the form of specially instantiated belief networks called Plan Recognition Networks (PRNs). Our initial experiments test several hypotheses concerning coordination performance as a factor of plan-recognition model expressiveness compared to coordination based upon using communication protocols of varying expressiveness. The initial results demonstrate that plan-recognition capabilities permit agents to realize better coordination performance in many situations due to increased knowledge provided through observation and inference compared to that supplied by communication protocols.