This is the download page for Bonet and Geffner's software GPT.

 

 

GPT is a planner with incomplete information, sensing and probabilities. The basic ideas used in the planner are discussed in the papers :

1. Planning with Incomplete Information as Heuristic Search in Belief Space.(Ps)
Blai Bonet and Hector Geffner. Proc. of 5th Int. Conf. on AI Planning and Scheduling. Breckenridge, CO. 2000. AAAI Press. Pages 52-61. 
2. GPT: A Tool for Planning with Uncertainty and Partial Information. (Pdf)
Blai Bonet and Hector Geffner. Workshop on Planning with Uncertainty and Partial Information, International Joint Conf. of AI. Seattle, WA. 2001. Pages 82-87.

The first paper discusses the various mathematical models and different algorithms used in the research. The second paper gives more details about the software, the implementation and syntax for inputting various domains etc.

The actual source code for the software is here : 

gpt-1.40-src-linux-080602.tar.gz

All the instructions for installing and running are in the README file in the package. Its an easy install and takes a couple of minutes for installation.

Useful tips on the use of GPT :

1. It solves conformant planning, contingent planning and probabilistic contingent planning problems. The output of comformant planning problems is a text file which has the plan besides other statistical data.
2. The output of contingent planning and probabilistic contingent planning is a gml file which can be viewed using bin/viewer program. The format of the viewed file is :
  -- Cyan : start state
  -- Yellow : Goal state(s)
  -- Red : Other intermediate states
  -- Labels on the states are actions the agent would perform in that state.
  -- Labels on the arrows are observations.
   One can use the mouse to rearrange the various nodes in the graph for easy viewing. You might need to zoom in for better view.
3. The examples in the aips-paper directory work fine. However, the distribution has various other examples of domains without a proper script.gpt file for them. Use the following sample script.gpt file. Note that problemname and domainname match with those in the pddlfilename. You could have a look at other features in these script files from examples/aips-paper/omelette/script.gpt.

parse problemname pddlfilename
compile probname_domainname
solve probname_domainname
generate graph probname.gml
generate table probname.table
shell make -f Makefile.probname_domainname deep-clean
quit