Minutes

These are the Minutes of the First PRECiS Domain Development Group Meeting held at the ARPI Review Meeting in Tucson Arizona on 24th February, 1994.

This was the first meeting of the PRECiS Domain Development Group. Previous discussions had taken place via e-mail and document interchange. The meeting was begun by each member present expressing their interest and possible use of the PRECiS materials. Those present and their affiliations were as follows:

The views and points of interest expressed by the members present were as follows:

• Matt Ginsberg Matt felt that the data contained in the current PRECiS materials was inconsistent and that the data presently was better suited to scheduling problems. As such the PRECiS materials should be changed to make then more suitable for generative planners. To this end Matt suggested removing the Appendices of the current document and making Version 2 of the PRECiS document contain only a single problem against which all groups interested in generative planning can carry out experiments.
• Lou Hoebel Felt the PRECiS simulator would be useful as such a suitable simulator seems to be missing from the Initiative. However, he felt that such a simulator need not be limited to logistics and that such simulators would help differentiate generative planners.
• Carla Ludlow Has made no use of the PRECiS materials but is in the process of creating a set of planning benchmarks for which PRECiS may prove useful.
• Yolanda Gil Yolanda's work is in the area of judging plans based on domain dependent criteria and the PRECiS simulator would help show whether the plan her system suggested was in fact the best.
• Brian Drabble and Austin Tate described the use of the PRECiS domain and data with the O-Plan2 system for generative plan purposes. The same data was being used by Glen Reece on an associated ASSERT funded Ph.D. to investigate issues related to Reactive Execution Agents.
• Others Martha Pollock (not present) Austin reported that Martha Pollock has a Pacifica based version of the Truckworld simulator and that Nathanial Martin at the University of Rochester is building a TRAINS world simulator based on Pacifica.

The general view of the meeting was that the scenarios in the Appendices of the document (as a demonstration domain) were unsuitable and that the generative planning community needed a single problem which every interested group could ``buy into''. The problem would have to be simple enough so that all generative planners at some level of abstraction, could solve it. The problem domain should be extended only in those cases where two or more groups required it. This would probably be the case in issues such as hierarchical modelling, variable handling, resources, authority, probabilities, etc. It was suggested by Matt Ginsberg that the scope of the problem should have a branching factor between 4 and 8 and require a search depth of 8 and 32. Austin felt that much larger plans should be allowed to test hierarchical planning properly. There was some discussion as to whether such a domain could be developed and it was felt that this issue should be further discussed by the exchange of e-mail. Matt agreed to advertise this and to coordinate the replies from interested groups.

After discussion it was decided that Version 2 of the PRECiS document would contain a description of the PRECiS /Pacifica domain and the series of data tables which define facts such as distances, transport capacities, transport ranges, etc. A seperate scenario document would contain a a new generative planning evaluation problem. The existing appendices of Version 1 of the PRECiS document would be kept as separate documents for possible use by other ARPI participants. The specific nature of the new scenario and the operators would be discussed via e-mail. It was felt that by having such a scalable generative planning domain that people in other related areas could use this information, e.g. Database design, COA generation and plan evaluation criteria.

The second issue which was addressed at the meeting was that of the use of domain simulators. Each of the groups present was agreed on their importance and a number of existing and future simulators for the PRECiS world have been identified. The current simulators are being developed at AIAI, University of Pittsburg (Martha Pollock) and at Rochester University where Nathanial Martin is developing a PRECiS simulator based on the TRAINS system.

Further discussions were undertaken to determine the relationship between data in the PRECiS domain and that in the IFD-2 and IFD-3 demonstrators. The mapping is sufficiently close (in terms of ports, airfields, major cities, etc) that should the need arise then the PRECiS domain could be converted to either of IFD-2 or IFD-3.