Criticality Metrics for Distributed Plan and Schedule Management

We address the problem of coordinating the plans and schedules for a team of agents in an uncertain and dynamic environment. Bounded rationality, bounded communication, subjectivity and distribution make it extremely challenging to find effective strategies. The Criticality-Sensitive Coordination (CSC) system uses multiple policy modification managers making predictable policy changes based on criticality metrics derived from simple computations on a graphrepresentation of the reward function with nearest neighbor communication. In the context of the DARPA Coordinators program, under an extensive and independent evaluation, the CSC system significantly outperformed competing approaches based on Temporal Networks and Markov Decision Processes.

• Type: Conference
• Year: 2008
• Where: AIPS
• Authors: Rajiv T. Maheswaran, Pedro A. Szekely
• Keywords: AIPS 2008, Artificial Intelligence, Bounded Rationality, Nearest Neighbor Communication, Predictable Policy Changes