Generating plans in concurrent, probabilistic, over-subscribed domains

  • Authors:

  • Li Li;Nilufer Onder

  • Affiliations:

  • Department of Computer Science, Michigan Technological University, Houghton, MI;Department of Computer Science, Michigan Technological University, Houghton, MI

  • Venue:
  • AAAI'08 Proceedings of the 23rd national conference on Artificial intelligence - Volume 2
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Planning in realistic domains involves reasoning under uncertainty, operating under time and resource constraints, and finding the optimal set of goals to be achieved. In this paper, we provide an AO* based algorithm that can deal with durative actions, concurrent execution, over-subscribed goals, and probabilistic outcomes in a unified way. We explore plan optimization by introducing two novel aspects to the model. First, we introduce parallel steps that serve the same goal and increase the probability of success in addition to parallel steps that serve different goals and decrease execution time. Second, we introduce plan steps to terminate concurrent steps that are no longer useful so that resources can be conserved. Our algorithm called CPOAO* (Concurrent, Probabilistic, Oversubscription AO*) can deal with the aforementioned extensions and relies on the AO* framework to reduce the size of the search space using informative heuristic functions. We describe our framework, implementation, the heuristic functions we use, the experimental results, and potential research on heuristics that can further reduce the size of search space.