The Gourmand Planner
Description
Gourmand is a solver for finite-horizon MDPs with large branching factors, built by Andrey Kolobov, Mausam, and Dan Weld. It is a further development of Glutton, the runner-up in the 2011 International Probabilistic Planning Competition (IPPC-2011) . The main difference between the two is their mode of operation: Gourmand is an online planner, while Glutton is an offline one. Gourmand has been shown to outperform both Glutton and PROST, the IPPC-2011 winner, across the set of all IPPC-2011 benchmarks.
Source code
A Gourmand implementation for Linux is available as part of G-pack. By downloading, using, or redistributing this software or its parts in any way, you automatically accept its license agreement.
For instructions on how to compile and run Gourmand, please see the README file included with G-pack's source code.
Publications about Gourmand and its versions
-
LRTDP vs. UCT for Online Probabilistic Planning.
Andrey Kolobov, Mausam, and Daniel Weld. AAAI'12.
PDF -
Reverse Iterative Deepening for Finite-Horizon MDPs with Large Branching Factors.
Andrey Kolobov, Peng Dai, Mausam, and Daniel Weld. ICAPS'12.
PDF
Relevant links
- Glutton, the precursor of Gourmand.
- PROST, a planning system by Thomas Keller and Patrick Eyerich from the University of Freiburg that won IPPC-2011.
- rddlsim, a toolkit by Scott Sanner that contains an implementation of the IPPC-2011 problem server and other useful utilities.
- miniGPT, a software package by Blai Bonet and Hector Geffner implementing several older planners, including LRTDP. G-pack borrows some of the code from it.