2 :0 0 p m - 6 :0 0 p m
TUTORIAL VIII
Agent Teams
Milind Tambe
University of Southern California

Audience: Researchers, educators, practitioners, and graduate students with an interest in intel-ligent agents will benefit from this tutorial.

Course Description: Teamwork is a critical capability in a large number of multi-agent environments. For instance, in virtual environ-ments for training and education, teamwork among agents is critical to provide the right environment for human trainees. Teamwork among software agents is crucial in applications that involve assistance to human organizations, as well as in information gathering, planning and logistics. Teamwork may also offer new methods for integrating heterogeneous software components. Finally, teamwork is obviously critical in distributed robotic applications as in future multi-spacecraft missions. In this tutorial, I will survey the state of the art in agent teamwork. Fundamentally, this area has focused on enabling different autonomous entities (e.g., software agents or robots) to work together, using inspirations from human teamwork. Thus, researchers have been attempting to fundamentally understand the nature of teamwork and practical techniques to rapidly construct robust agent teams. One key lesson learned is that in complex, uncertain environments, creating fixed, domain-specific coordination plans to manage teamwork is highly problem-atic: these plans are not reusable across domains, and their lack of flexibility can lead to severe failures. Instead, a new approach based on developing a general teamwork model appears to provide more promise. These models enables agents to explicitly reason about commitments and responsibilities in teamwork, and flexibly plan own coordina-tion. Furthermore, these models are reusable across domains, aiding rapid construction of agent teams. A key outcome of this work is the notion of "team-oriented programming" to enable developers to program agent teams at a high-level, while the coordination is automatically generated at run time due to the teamwork model.

Building on a similar tutorial delivered at the European Summer School on Agents, and classes on multi-agents taught at the University of Southern California's computer science department, this tutorial will first cover theory of teamwork and then practical teamwork models based on such theory. We will cover one particular teamwork model, called STEAM, in more detail. STEAM has been the basis of several agent teams that we have developed, including pilot teams for battlefield simulations, top-performing agent teams for RoboCup soccer, teams integrating software agents over the internet, and more recently, "Electric Elves", a team of software assistants that help researchers in their daily activities, operating 24/7 at ISI since 6/1/2000.

Lecturer: Milind Tambe is an Associate Professor of Computer Science at the University of Southern California (USC) and a project leader at USC's Information Sciences Institute (ISI). His research interests are in the areas of multi-agent systems, particularly, in topics such as teamwork, coordination, and negotiations and agent modeling. A member of the board of directors of the International Foundation for Multi-agent Systems, and a trustee of the RoboCup Federation, he is also on the editorial board of the Journal of Autonomous Agents and Multi-agent Systems, Journal of Artificial Intelligence Research, and IEEE Intelligent Systems. He has served as the program co-chair for the International Conference on Multi-agent Systems (ICMAS'2000), and as senior program committee member for the AAAI and Agents conferences. (http://www.isi.edu/team-core/ tambe)