Team Leader: An Approach to Mixed-Initiative Agent Team Management and Evaluation

Abstract

Abstract : The rapid growth in research and development of agent-based software systems has led to concerns about how human users will control the activities of teams of agents that must actively collaborate. Practical multi-agent systems will often be comprised of small teams of heterogeneous agents, under direct supervision by users acting as team leaders. This research focused on a number of areas critical to tasking and managing teams of humans and agents, including techniques for agentizing' legacy systems, Human-Computer Interaction techniques for interactively defining tasks and roles and communications between agents, mechanisms for interactively forming and controlling teams of agents as well as summarizing and visualizing agent task status, methods for translating information across different ontologies, and mechanisms for monitoring and controlling agent communications based on semantic content. A key contribution of this work has been the development of an approach to dynamic information sharing among agents on teams, alleviating human managers from the need to coordinate information awareness. The work was demonstrated in a number of large-scale demonstrations including the Coalition Agents experiment (CoAX) and the Mixed-Initiative Agent Team Administration (MIATA) demonstration.