Towards Heterogeneous Agent Teams

Abstract

Agent integration architectures enable a heterogeneous, distributed setof agents to work together to address problems of greater complexitythan those addressed by the individual agents themselves. Unfortunately,integrating software agents and humans to perform real-world tasks in alarge-scale system remains difficult, especially due to two keychallenges: ensuring robust execution in the face of a dynamicenvironment and providing abstract task specifications without all thelow-level coordination details. To address these challenges, ourTeamcore project provides the integration architecture withgeneral-purpose teamwork coordination capabilities. We make each agentteam-ready by providing it with a proxy capable of general teamworkreasoning. Thus, a key novelty and strength of our framework is thatpowerful teamwork capabilities are built into its foundations byproviding the proxies themselves with a teamwork model called STEAM.While STEAM has earlier been demonstrated in domains involvinghomogeneous agent teams, its use in Teamcore proxies illustrates thatteamwork models may also be applied in domains involving heterogeneousagents. Given STEAM, the Teamcore proxies addresses the first agentintegration challenge, robust execution, by automatically generating therequired coordination actions for the agents they represent. We can alsoexploit the proxies' reusable general teamwork knowledge to address thesecond agent integration challenge. Through team-oriented programming, adeveloper specifies a hierarchical organization and its goals and plans,abstracting away from coordination details. Our integration architectureenables teamwork among agents with no coordination capabilities, and itestablishes and automates consistent teamwork among agents with somecoordination capabilities. We illustrate how the Teamcore architecturesuccessfully addressed the challenges of agent integration in twoapplication domains: simulated rehearsal of a military evacuationmission and facilitation of human collaboration.