Probabilistic Graphplan (PGP/TGP) Home Page

School of Computer Science, Carnegie Mellon University, Pittsburgh PA 15213-3891
People: Avrim Blum and John Langford

PGraphplan and TGraphplan are Graphplan-based planners for STRIPS-style domains that include probabilistic actions. Given a start state, a time horizon, and a set of goals, PGraphplan finds the contingent plan with highest probability of success within the horizon. It does this by performing a standard "top-down dynamic programming" approach, but using the planning graph to constrain the search space. TGraphplan finds potentially sub-optimal plans, but in general runs much more quickly than PGraphplan, and uses a search much closer in spirit to the original Graphplan algorithm. This web page contains pointers to the ECP paper describing these planners, and to code and domains for those who with to try them out.

Source code and domains:
The domain format should be clear from the examples, but feel free to send us email if you are having trouble creating your own. Code for TGraphplan should be here soon.

Avrim Blum and John Langford, Probabilistic Planning in the Graphplan Framework. In the 5th European Conference on Planning (ECP'99).
(c) Springer-Verlag.

See also
Graphplan home page.

Algorithms and Complexity | Computer Science Department | School of Computer Science

This page maintained by Avrim Blum (