CAST: Collaborative Agents for Simulating Teamwork

Abstract

Psychological studies on teamwork have shown that an effective team often can anticipate information needs of teammates based on a shared mental model. Existing multi-agent models for teamwork are limited in their ability to support proactive information exchange among teammates. To address this issue, we have developed and implemented a multi-agent architecture called CAST that simulates teamwork and supports proactive information exchange in a dynamic environment. We present a formal model for proactive information exchange. Knowledge regarding the structure and process of a team is described in a language called MALLET. Beliefs about shared team processes and their states are represented using Petri Nets. Based on this model, CAST agents offer information proactively to those who might need it using an algorithm called DIARG. Empirical evaluations using a multi-agent synthetic testbed application indicate that CAST enhances the effectiveness of teamwork among agents without sacrificing a high cost for communications.