Towards any-team coaching in adversarial domains

As multi-agent systems continue to grow in importance, the types of relationships between agents continue to be studied. One important relationship that humans often exhibit is a coach. For example, the lead programmer in a software development team provides structure, direction, and a problem decomposition to the other programmers and a professor provides guidance and advice to her graduate students in search of their Ph.Ds. In order to explain the problem a coach faces, one must first define what the role of a coach is. A coach is a member of a team in the sense of having a common goal. However, in common usage (such as used in sports), a distinction is usually made between the coach and the team of players. We preserve this common usage here, and discuss a single coach working with multiple teams of agents. Unlike other agents in the team, the coach’s only action is to communicate to the agents on the team, which we will call the receivers. The coach’s goal is to improve the performance of the team through this communication. The communications from the coach should suggest changes to the receivers’ behavior. The expressiveness and flexibility of communication languages can vary greatly. Advice can be very specific, such as “In this state, take this action” or very general, such as “Your goal should now be this.” Also, for more general advice, the coach may want the agents to be independent and not follow advice in all situations. Given the explanation of the coach role, the coaching problem can be stated quite simply: “How can an agent in

[1]  Andreas Birk,et al.  RoboCup 2001: Robot Soccer World Cup V , 2002, Lecture Notes in Computer Science.

[2]  Ian Frank,et al.  Soccer Server: A Tool for Research on Multiagent Systems , 1998, Appl. Artif. Intell..

[3]  Milind Tambe,et al.  Adjustable autonomy in real-world multi-agent environments , 2001, AGENTS '01.

[4]  Manuela Veloso,et al.  An Empirical Study of Coaching , 2002, DARS.