Overview SPUDD stands for "Stochastic Planning using Decision Diagrams". This page is intended to provide information and access to the SPUDD module that was implemented in the department of Computer Science   at the   University of British Columbia . SPUDD implements a value iteration algorithm for MDPs and POMDPs that uses algebraic decision diagrams (ADDs) to represent value functions and policies.

People Jesse Hoey Robert St-Aubin Alan Hu Craig Boutilier

Spudd online For small problems if you want to test out how SPUDD works, try submitting your own MDP through Interactive SPUDD Spudd online is now Spudd offline. Permanently (for a while at least until I find a place to host it). Instead, download the code below and run it yourself!

Downloads NEWMarch 1st, 2007. spudd version 3.5.3. fixed two minor problems to get Spudd to compile with gcc v4. April 13th, 2005. spudd version 3.5.2. A minor bug was introduced in version 3.5, which is now removed. This bug caused value iteration to converge much more slowly, and also affected problems with state-dependent action costs. Note: the POMDP solution code has problems with caching (I think), causing it to run very slowly on larger problems. Contact the author for further details. February 16th, 2005 spudd version 3.5. Improved POMDP support (point-based algorithm). gcc 3.4 compatible. This version was compromised. Please upgrade to version 3.5.2 if you were using 3.5 September 10, 2004 spudd version 3.4.1 small bug fixes - POMDP solving using Fast Point Based Value Iteration works. The default Spudd binary that is built will be normal MDP Spudd, so current users should see no difference. Contact Jesse Hoey for more details. For older versions, go here This code is for research purposes only. Please see the README file in Spudd/doc/ for instructions

Publications Optimal and Approximate Stochastic Planning using Decision Diagrams Jesse Hoey , Robert St-Aubin, Alan Hu and Craig Boutilier University of British Columbia Technical Report TR-00-05, June 2000 Complete description and results for both optimal and approximate algorithms. you can read the abstract or retrieve the paper ( gzipped ps 143K). SPUDD: Stochastic Planning using Decision Diagrams Jesse Hoey , Robert St-Aubin, Alan Hu and Craig Boutilier In Proceedings of the Fifteenth Conference on Uncertainty in Artificial Intelligence (UAI 99), Stockholm, Sweden The original paper describing only the optimal version SPUDD. you can read the abstract or retrieve the paper ( gzipped ps 69K). APRICODD: Approximate Policy Construction using Decision Diagrams Robert St-Aubin, Jesse Hoey and Craig Boutilier in Advances in Neural Information Processing 13 ( NIPS 2000). you can read the abstract or retrieve the paper ( gzipped ps 43K or pdf 81K).

Credits Many people have helped by finding bugs and giving valuable advice or suggestions Mark Plutowski, Carlos Guestrin, Stephen North, Li Li, Jen Boger, Nigel Morris, Pascal Poupart, Yaxin Liu, Rikin Gandhi, Liam Stewart, Alberto Reyes,... Please send your comments or feedback to the .