Depth-first iterative-deepening: an optimal admissible tree search
Artificial Intelligence
Sokoban: Evaluating Standard Single-Agent Search Techniques in the Presence of Deadlock
AI '98 Proceedings of the 12th Biennial Conference of the Canadian Society for Computational Studies of Intelligence on Advances in Artificial Intelligence
Finding optimal solutions to Rubik's cube using pattern databases
AAAI'97/IAAI'97 Proceedings of the fourteenth national conference on artificial intelligence and ninth conference on Innovative applications of artificial intelligence
AAAI'96 Proceedings of the thirteenth national conference on Artificial intelligence - Volume 1
Computer Go: an AI oriented survey
Artificial Intelligence
A Performance Analysis of Transposition-Table-Driven Work Scheduling in Distributed Search
IEEE Transactions on Parallel and Distributed Systems
Relevance Cuts: Localizing the Search
CG '98 Proceedings of the First International Conference on Computers and Games
Domain-dependent single-agent search enhancements
IJCAI'99 Proceedings of the 16th international joint conference on Artifical intelligence - Volume 1
| Hi-index | 0.00 |
Single-agent search is a powerful tool for solving a variety of applications. Most of the application domains used to explore single-agent search techniques have the property that if you start with a solvable state, at no time in the search can you reach a state that is unsolvable. In this paper we address the implications that arise when state transitions can lead to unsolvable (deadlock) states. Deadlock states are partially responsible for the failure of our attempts to solve positions in the game of Sokoban. In this paper, we introduce pattern search, a real-time learning algorithm that identifies the minimal conditions (pattern) necessary for a deadlock, and applies that knowledge to eliminate provably irrelevant parts of the search tree. Identification of deadlock patterns is equivalent to correcting the heuristic lower bound of a position to infinity. Generalizing pattern searches to find arbitrary lower bound increases yields a powerful new search enhancement. In the game of Sokoban, pattern searches result in a 15-fold reduction of the cost of each additional IDA* iteration.