ADAPTIVE AGENT TRACKING IN REAL-WORLD MULTIAGENT DOMAINS - A PRELIMINARY-REPORT

Intelligent interaction in multi-agent domains frequently requires an agent to track other agents' mental states: their current goals, belie fs and intentions. Accuracy in this agent-tracking task is critically dependent on the accuracy of the tracker's (tracking agent's) model of the trackee (tracked agent). Unfortunately, in real-world situations, model imperfections arise due to the tracker's resource and informati on constraints, as well as due to trackees' dynamic behavior modificat ion. While such model imperfections are unavoidable, a tracker must no netheless attempt to be adaptive in its agent tracking. This article i dentifies key issues in adaptive agent tracking and presents an approa ch called DEFT. At its core, DEFT is based on discrimination-based lea rning. The main idea is to identify the deficiency of a model based on tracking failures, and revise the model by using features that are cr itical in discriminating successful and failed tracking episodes. Beca use in real-world situations the set of candidate discriminating featu res is very large, DEFT relies on knowledge-based focusing to limit th e discrimination to those features that it determines were relevant in successful tracking episodes-with an autonomous explanation capabilit y as a major source of this knowledge. This article reports on experim ents with an implementation of key aspects of DEFT in a complex synthe tic air-to-air combat domain. (C) 1998 Academic Press Limited.