• Sokoban and its search-space properties are investigated and shown to be a significant challenge for current AI research.
  • The concept of macro moves is improved by adding automatic off-line macro-move generation.
  • A new search enhancement is introduced: Pattern searches. Pattern searches gather knowledge dynamically.
  • A new forward pruning technique is introduced: Relevance cuts.
  • A new, knowledge-based overestimation technique is proposed.
  • A classification of enhancements is given.
  • New concepts of task and control knowledge are proposed.
  • A framework for single-agent search enhancements is given.

GPW'99, October 16, 1999. Pushing the Limits: New Developments in Single-Agent Search previous up next