Using Abstraction for Heuristic Search and Planning

Abstract

My research interests are in the areas of planning, computer games, and heuristic search. My objectives include using abstraction to solve hard heuristic search problems. Classical heuristic search has been successful for games like Chess and Checkers, but seems to be of limited value in games such as Go and Shogi, and puzzles such as Sokoban. My work is focused on developing a planning approach for Sokoban, aiming to overcome the limitations exhibited by current approaches. In Sokoban, a man in a maze has to push stones from their current location to designated goal locations. The game is difficult for a computer for several reasons including deadlocks (positions from which no goal state can be reached), large branching factor (can be over 100), long optimal solutions (can be over 600 moves), and an expensive lower-bound heuristic estimator, which limits search speed.