Event Location: 1305 Newell Simon Hall
Bio: Ed Durfee is a Professor of Computer Science and Engineering, and of Information, at the University of Michigan, where he has served on the faculty for over 20 years. His research focuses on developing representations and algorithms for multiagent planning, scheduling, and coordination, with applications that include cooperative robotics, service-oriented computing, and cognitive orthotics. He received his AB degree from Harvard University and his PhD from the University of Massachusetts.

Abstract: Effective coordination among cooperating agents typically improves with greater mutual awareness. However, the costs and delays in achieving and maintaining mutual awareness, and reasoning about detailed models of others’ beliefs, goals, and plans, can impede the responsiveness and effectiveness of a multiagent system. That is, cooperating agents can arguably benefit from intentionally knowing less about each other, and revealing less about themselves. In this talk, I describe some of our ongoing research that uses abstraction and conditional decoupling to achieve effective coordination without excessive overhead and without unnecessarily overconstraining individuals’ actions. Specifically, I summarize abstraction techniques that can dramatically accelerate search for optimal joint policies in decentralized partially-observable Markov decision processes, and distributed decoupling algorithms that can support privacy and flexibility in multiagent scheduling problems.