Sokoban: A Challenging Single-Agent Search Problem (1997)
| Venue: | In IJCAI Workshop on Using Games as an Experimental Testbed for AI Reasearch |
| Citations: | 12 - 1 self |
BibTeX
@INPROCEEDINGS{Junghanns97sokoban:a,
author = {Andreas Junghanns and Jonathan Schaeffer},
title = {Sokoban: A Challenging Single-Agent Search Problem},
booktitle = {In IJCAI Workshop on Using Games as an Experimental Testbed for AI Reasearch},
year = {1997},
pages = {27--36},
publisher = {Universiteit}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Sokoban is a challenging single-player game -- for both man and machine. The simplicity of the rules belies the complexity of the game. This paper describes our program, Rolling Stone, a first attempt to solve Sokoban problems. Adapting and extending the standard single-agent search techniques in the literature, we are able to optimally solve 12 of 90 problems from a standard test suite. This result demonstrates how difficult a game Sokoban really is for computers to solve, and underlines the need for more sophisticated search techniques, including planning. Keywords: single agent search, heuristic search, Sokoban, macro moves, deadlocks 1 Introduction Sokoban is a popular one-player computer game. The game apparently originated in Japan, although the original author is unknown. The game's appeal comes from the simplicity of the rules and the intellectual challenge offered by deceptively easy problems. The rules of the game are quite simple. Figure 1 shows a sample Sokoban problem ...
Keyphrases
challenging single-agent search problem first attempt popular one-player computer game standard single-agent search technique macro move original author sophisticated search technique intellectual challenge easy problem single agent search sokoban problem rolling stone game sokoban sample sokoban problem introduction sokoban heuristic search standard test suite single-player game
