Creating the First International Probabilistic Planning Competition

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

This small planning grant will support efforts to launch the first international probabilistic planning competition, slated to be held in conjunction with the 2004 International Conference on Automated Planning and Scheduling (ICAPS-04). The competition is being organized as a parallel track of the Fourth International Planning Competition (IPC) at ICAPS-04. The IPC was first held in 1998 in conjunction with the Artificial Intelligence Planning and Scheduling conference (AIPS). Follow-up competitions were held in 2000 and 2002, and it is widely believed that the competitions have been enormously valuable to the planning community, spurring advances in algorithms and representations and rapidly disseminating them throughout the field. The organizers of ICAPS-04 will be hosting the 4th IPC, which has focused exclusively on classical planning, involving sequential decision making when the effects of all decisions are deterministic. Over the past decade, as computers have become more and more integrated with the physical world, researchers have begun to study models that can represent the inherent uncertainty present in the effects of many decisions. In this setting, plans are evaluated by how likely they are to achieve desired outcomes. The dominant mathematical model in this community is the Markov decision process (MDP) model, and there are research groups all over the world studying MDPs for planning. However, progress in the field is inhibited by the lack of a unified metric for progress and little sharing of results between groups. A probabilistic planning competition could give the probabilistic planning community a much needed opportunity to come together and move forward. This award will support adaptation of a plan validator program and travel to facilitate organizing of the competition. Its broader impacts include training a graduate student and improvement of computerized planning systems to plan many kinds of real-world activities.