From FOND to Probabilistic Planning: Guiding search for quality policies. Camacho, A., Muise, C., Ganeshen, A., & McIlraith, S. A. In Workshop on Heuristic Search and Domain Independent Planning (HSDIP'15), 2015.
From FOND to Probabilistic Planning: Guiding search for quality policies [pdf]Paper  abstract   bibtex   1 download  
We address the class of probabilistic planning problems where the objective is to maximize the probability of reaching a prescribed goal (MAXPROB). State-of-the-art probabilistic planners, and in particular MAXPROB planners, offer few guarantees with respect to the quality or optimality of the solutions that they find. The complexity of MAXPROB problems makes it difficult to compute high quality solutions for big problems, and existing algorithms either do not scale well, or provide poor quality solutions. We exploit core similarities between probabilistic and fully observable non-deterministic (FOND) planning models to extend the state-of-the-art FOND planner, PRP, to be a sound and sometimes complete MAXPROB solver that is guaranteed to sidestep avoidable deadends. We evaluate our planner, ProbPRP, on a selection of benchmarks used in past probabilistic planning competitions. The results show that ProbPRP outperforms previous state-of-the-art algorithms for solving MAXPROB, and computes substantially more robust policies, at times doing so orders of magnitude faster.
Downloads: 1
About
Documentation
Keywords
Search
Users
Terms and Conditions of Use
Privacy Policy
Sitemap
© BibBase.org 2021