Welcome to the website of the probabilistic planning system PROST, which participated at the (fully observable) MDP track of IPPC 2011 and IPPC 2014 and won both competitions. An overview on the algorithms used in the IPPC 2011 configuration is described in the ICAPS 2012 paper: PROST is based on the UCT algorithm, which is initialized with a iterative deepening search on a most-likely determinized simplification of the original task. The used input language is RDDL, a brief documentation of which can be found here. If you want to cite the PROST planner, this is the correct paper to do so, even though the latest version contains only few lines of code that ran already in 2011.
In 2013, the PROST planner has been extended and is now a framework that supports several UCT variants which are described in the ICAPS 2013 paper on the Trial-based Heuristic Tree Search (THTS) framework, a generalization of the well-known Monte-Carlo Tree Search framework. If you mention the algorithms that are implemented in PROST, this is most likely the correct paper to cite. An overview on the implemented algorithms is given in Thomas' PhD thesis, where several anytime optimal THTS algorithms have been implemented in PROST and compared.
In case you have any questions about the usage of PROST, feel free to write us an email. If you encounter a bug, please check the issue-tracker and add your issue if it is not already there or leave a comment in the existing issue to let us know that the feature or bug is important to your work.
- We have added a new component to the planner that reduces its dependency from rddlsim: from now on, it is possible to directly parse rddl files (without having to translate them to rddl_prefix first). Even though we both parsers can be used at the moment, the rddl_prefix parser will not be maintained any longer (and will be removed eventually). (17 October 2016)
- We moved the website and repository to be hosted at bitbucket, which provides us with an an affiliation-independent platform and also some new features like this Wiki or a proper issue-tracker. (26 August 2016)
- Over the past few months, we have altered the planner structure significantly. It closely resembles the THTS ingredient structure with own classes for each ingredient. Most importantly, this allows to specify a large number of THTS algorithms by simply mixing and matching the corresponding ingredients. (21 December 2015)
- Winner of IPPC2014! The IPPC 2014 results are published and we have successfully defended our title! (27 June 2014)
- The planner has been extended to work with finite-domain variables instead of boolean ones, which allows the usage of enums (Issue
#4). (20 February 2014)
- The THTS framework and the UCT variants described have been merged into the PROST framework (Issue
#6). (29 May 2013)
- Our paper Trial-based Heuristic Tree Search for Finite Horizon MDPs, a collaboration with Malte Helmert and with experiments implemented in PROST, has been awarded the Best Student Paper Award of the 23rd International Conference on Automated Planning and Scheduling (ICAPS 2013). (20 May 2013)
- State action constraint handling has been added to the framework (Issue
#2). For the first time, PROST supports a larger fragment of RDDL than the one that was used for IPPC 2011. (1 April 2013)
- The old website "prost.informatik.uni-freiburg.de" which now redirects here went online. (27 August 2012)
- The PROST planner is the Winner of IPPC 2011. (14 June 2011)
There are quite a few people who have contributed to PROST since the planner's development was started around December 2010, and this is the place where we give credit to whom credit is due (these are ordered alphabetically):
- Patrick Eyerich (core developer, 2010-2012)
- Florian Geißer (core developer, 2014-2016)
- Max Grüner (bug fix, 2015)
- Thomas Keller (core developer, 2010-2016)
- Dorde Relic (RDDL parser, 2016)
- Scott Sanner (TCP/IP client, 2010)
- Joris Scharpff (readme and installation description, 2013)
- Florent Teichteil-Koenigsbuch (Windows support, 2016)
Finally, if you
- have contributed to PROST as well and your name is not in this list (we apologize sincerely in this case, there surely was no harm intended!),
- have implemented some algorithm or feature in the PROST framework that you would like to see published,
- wish to collaborate scientifically, or
- want to contribute to the further development of PROST
please to let us know via email!