Conclusions
  • Sokoban is a challenge significantly more difficult than domains traditionally used for research.
  • Beware of unqualified research results using "simple" domains!


  • Textbook approaches are nice and clean.
  • Current state-of-the-art requires domain-dependent knowledge.
  • Practice is complex and messy, lots of engineering needed.


  • The more specific the knowledge, the more powerful it is!
  • Search uncovers valuable, reusable knowledge that pertains to specific situations:
    Exceptions, special cases, configurations never seen before.


  • Simplify the design, development and test by splitting the problem.
  • Differentiate between task and control knowledge.

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