Andreas Junghanns and Jonathan Schae#er. Sokoban: A case-study in the application of domain knowledge in general search enhancements to increase e#ciency in single-agent search. Artificial Intelligence, special issue on search, 2000.  Home/Search   Document Details and Download   Summary   Related Articles

....2 4 18 0 50 12 40 E#ective 2. 3 13.3 10 7 Solution length Typical 100 18 260 45 Range 1 112 1 20 97 674 8 70 Search space size Upper bound 10 Underlying graph Undirected Undirected Directed Directed Table 1: Search space properties of di#erent games (adapted from Junghanns [JS00]; additional sources are [KT96, Kor97, EK98] The e#ective branching factor is the number of children of a state, after applying memory bounded pruning methods (in particular, not utilizing transposition tables; see Sec. 5.3.1 for the methods applied to Atomix) For Sokoban and Atomix, the ....

....are quite long (100 600 moves) and the branching factor can be even larger than for Atomix. Common with Atomix is the di#culty in isolating subgoals. The best known admissible heuristic for Sokoban needs to perform minimum cost perfect matching to assign stones to goal positions optimally [JS00]. It takes O(#stones ) time to calculate per state, even when reusing information from the parent state. Therefore, much less states can be explored; while for Atomix 1,000,000 states per second can be generated, this number is for Sokoban around 10,000. Sokoban has been shown to be ....

[Article contains additional citation context not shown here]

Andreas Junghanns and Jonathan Schae#er. Sokoban: A case-study in the application of domain knowledge in general search enhancements to increase e#ciency in single-agent search. Artificial Intelligence, special issue on search, 2000.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC