Plan compilation home page

Traditional AI planners have typically done very little compilation of a planning problem. Recently, several techniques for compiling planning problems have been developed and applied with great success. At the AIPS 98 planning conference every competing planner did some form of plan compilation. However, all of these planners used different compilation techniques. The goal of this web page is to work as a central repository for these compiling planners and promote sharing of the techniques used.

Here are several of the planners:

GraphPlan was the first compiling planner by Avrim Blum and Merrick Furst. New features included: compiling onto a graph, parallel operators, and propagating pairwise constraints through the graph. source: local, remote.
BlackBox by Henry Kautz and Bart Selman compiles onto a SAT representation then uses fast SAT solvers such as walksat to solve the planning problem. source: local, remote.
HSP by Blai Bonet and Hector Geffner automatically generates a heuristic function to bias search. For speed, the heuristic function is generated in C, compiled, and linked in before the search begins. source: local
IPP by Jana Koehler and several others has made many improvements to representation and extended the language for describing problems. source: local, remote.
Sensory Graphplan (SGP) by Corin Anderson and Dan Weld is a rewrite of graphplan in lisp with some extensions and modifications.
STAN by Maria Fox and Derek Long has improved the underlying graph representation, improved the set representation, and uses type inferencing methods to prune the search space.

Jana Koehler did some analysis of the results of the AIPS 98 planning competition.

jcl@cmu.edu