Knowledge-Based Human-Agent Teamwork for Distributed Training

摘要

This paper presents a knowledge-based approach to human-agent mixed teams for distributed team training in CAST-DDD. CAST-DDD is a marriage between the multi-agent architecture CAST and the command-and-control simulation tool DDD, where CAST agents can replace some or all members on a DDD team. We explore the MALLET language in CAST to capture DDD teamwork knowledge that involve both humans and agents. To allow for adjustable autonomy of human team members, we provide a repetitive choice construct for embedding non-deterministic operations in human-agent team processes. To offer a visible picture of teamwork status, team processes are visualized and tracked via an extended formalism of Predicate/Transition nets. In addition, we describe different communication and coordination methods for members of a human-agent mixed team as well as how agents reason about the dynamic, partially observable environment of the DDD simulation.