Plus d'informations ici
Coordination in large multiagent systems : do agents need stock options ?
Intervenant(s) : Pr. Kagan TUMER (Oregon State University)Coordinating large autonomous systems operating in dynamic and stochastic environments is imperative in many real world problems today. Air traffic management, in particular, is a great example of a domain where the need for tight coordination among heterogeneous agents is critical to safe, reliable and efficient performance. But many coordination approaches either focus on the emergent behavior of simple agents (e.g., swarms) or on the explicit coordination of a small set of agents (e.g., multi-robot coordination). In this talk, I will describe how to develop the objectives, utilities and incentives to coordinate large numbers of sophisticated agents. To borrow the terminology from internet boom, I will describe the equivalent of providing “stock options” for each agent, so that when each agent achieves its own objective, the system as a whole achieves the system designer’s objective. I will show results from robot coordination and air traffic management domains to highlight the applicability, scalability and robustness of this approach.
cedric.herpson (at) null