Implementing Shared Mental Models for Collaborative Teamwork

Psychology studies have shown that one of the keys to human teamwork is the ability of teammates to anticipate the needs of others and proactively take appropriate action using an overlapping shared mental model (SMM). This paper introduces an implemented multi-agent architecture – CAST, which enables a team of agents to establish a computational shared mental model. Using such a SMM, agents can (1) anticipate others’ information needs, (2) perform dynamic task allocations with minimum interagent negotiation, (3) choose whether to proactively inform teammates about the information relevant to their needs, and (4) rationally fuse information to meet teammates’ different levels of information needs. We evaluate our approach through two experiments in simulated battlefield domains. The studies show the vision of empowering agents with SMMs for proactive teamwork behaviors is both challenging and promising, which can benefit the development of agent-based solutions for training and supporting mixed human-agent teams that need to filter and fuse an overwhelming amount of information, and to make critical decisions under time constraints.