We propose to create novel planners for large, structured stochastic planning domains by leveraging determinization in a principled manner. Our methods exploit hindsight optimization and add innovative search and machine-learning techniques to address the weaknesses of that technique. We further discuss the computational efficiency issues raised by the need to solve many related deterministic planning problems, and propose to develop novel theoretical characterizations of the hindsight optimization technique.
|Effective start/end date||7/1/09 → 9/30/13|
- National Science Foundation (NSF): $328,821.00