The Value of Communication in Decentralized Planning and Control
Final rept. 2 Jan 2003-31 May 2006
MASSACHUSETTS UNIV AMHERST DEPT OF COMPUTER SCIENCE
Pagination or Media Count:
This project produced a mathematical approach to the design and analysis of decentralized control with particular focus on management of communication in an uncertain environment. Target applications of this framework involve multiple decision makers that could exchange a limited amount of information because communication is either costly or risky. The main accomplishments of the project include the formulation of a general mathematical model to study such problems called decentralized POMDP developing the first policy iteration algorithm for solving general problems formalized as decentralized POMDPs developing several efficient algorithms for solving special classes of decentralized POMDPs that arise in practice developing an effective myopic approach to communication in collaborative multi-agent systems and laying the foundations for developing learning techniques for communication. The report describes these research accomplishments and provides references to published papers and PhD dissertations that include detailed descriptions of the results.
- Information Science
- Operations Research