AAAI Publications, Sixth European Conference on Planning

Font Size: 
Beyond Plan Length: Heuristic Search Planning for Maximum Reward Problems
Jason Farquhar, Chris Harris

Last modified: 2014-05-21

Abstract


Automatic extraction of heuristic estimates has been extremely fruitful in classical planning domains. We present a simple extension to the heuristic extraction process from the well-known HSP and FF systems which allow us to apply them to reward maximisation problems. These extensions involve computing an estimate of the maximal reward obtainable from a given state when ignoring delete lists, which are then used to guide the backward search in the FF system. The heuristics are evaluated in a simple robotic delivery task and shown to be effective in reducing the number of nodes evaluated. In this way we seek to apply recent advances in classical planning to a broader range of problems.

Keywords


Domain Independent Planning, Reward Based Planning, Heuristic Search Planners

Full Text: PDF