[工程科技]PRODIGY40 The Manual and Tutorial 1 The PRODIGY Research Group_ under the supervision of J.pdf

PRODIGY4.0: The Manual and Tutorial1ThePRODIGYResearch Group, under the supervision of Jaime G. Carbonell: Jim Blythe, Oren Etzioni, Yolanda Gil, Robert Joseph, Dan Kahn, Craig Knoblock, Steven Minton, Alicia P´ erez, Scott Reilly, Manuela Veloso, and Xuemei Wang.2June 1992 CMU-CS-92-150School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213AbstractPRODIGYis a general-purpose problem-solving architecture that serves as a basis for research in planning,machine learning, apprentice-type knowledge-refinement interfaces, and expert systems. This document is a manual for the latest version of thePRODIGYsystem,PRODIGY4.0, and includes descriptions of thePRODIGYrepresentation language, control structure, user interface, abstraction module, and other features. The tutorial style is meant to provide the reader with the ability to runPRODIGYand make use of all the basic features, as well as gradually learning the more esoteric aspects ofPRODIGY4.0.1This research was sponsored by the Avionics Laboratory, Wright Research and Development Center, Aeronautical SystemsDivision (AFSC), U. S. Air Force, Wright-Patterson AFB, OH 45433-6543 under Contract F33615-90-C-1465, Arpa Order No. 7597. The views and conclusionscontained in this documentare those of the authors and should not be interpreted as representingthe official policies, either expressedor implied, of the U.S. Government.2TheinitialdesignwasfullydevelopedandimplementedbyJimBlythe,XuemeiWang,andDanKahn. TheauthorsacknowledgeSteven Minton, Jaime Carbonell, Craig Knoblock, and Dan Kuokka in designing and implementing the initialPRODIGYsystem,PRODIGY2.0. Manuela Veloso and Daniel Borrajo designed and implemented the first nonlinear problem solver forPRODIGY, NOLIMIT. Our past experiencewithPRODIGY2.0 and NOLIMITdrove our design and implementation ofPRODIGY4.0. Most of this manual was written by Jim Blythe, YolandaGil, Robert Joseph,Alicia P´ erez (editor), and Xuemei Wang.Keywords: planning, problem solving, machine learning, Prodigy1IntroductionPRODIGYis a domain-independent problem-solving architecture used primarily as a testbed for research in planning, machine learning, and knowledge acquisition. AlthoughPRODIGYcould be employed as a theorem prover or an expert system, it is primarily intended to be used as a general planner. In particular, thePRODIGYarchitecture is well suited for:1. A basic means-ends analysis nonlinear planner in high-level, symbolic domains.2. A tool for exploring the space of planners in complex and unusual domains (via domain-oriented control rules and other adaptations).3. Atestbedformachinelearningresearch, primarilyspeed-uplearning(EBL,analogy,abstraction,etc.), but also learning at the knowledge-level (experimentation), and interactive knowledge acquisition (see Appendix A for an overview of these mechanisms).Points two and three are important:PRODIGYprovides a platform for research in planning and learning that eliminatesthe need for each researcher tostart from scratch and designher or hisown planningsystem. AlthoughPRODIGYdoes not address certain issues (such as representing uncertainty), it can be used as an initial model, to aid in the design or serve as the basis for a more complex planner.PRODIGYis a powerful and well-engineered system. One of our main concerns is to makePRODIGYa usable system.PRODIGYhas been applied to many different domains: robotic path planning, the blocksworld, an augmented version of theSTRIPSdomain, multirobot task planning, matrix algebra multiplication, discretemachine-shop planning and scheduling, process planning, computer configuration, logistics planning, and several others.1.1PRODIGY4.0The current version of thePRODIGYproblem solver incorporates several features with respect to previous versions.PRODIGY4.0 is a nonlinear problem solver allowing full interleaving of goals. It works with a set of goals, as opposed to the top goal in a goal stack, asPRODIGY2.0 does. Improving the matcher efficiency hasbeen alsoamajoreffortofthe project. Themulti-levelabstractionplanningcapabilityisfullyintegrated withtheproblemsolver. In addition,PRODIGY4.0 providesinterruptionhandling,implementsa simpleform of dependency-directed backtracking, and allows multiple problem spaces.1.2How to Use This ManualSection 2 contains a very simple example of howPRODIGYsolves a problem.This section is a good introduction to how the system works. Its subsections introduce the rest of the manual. Section 3 showshow to represent domain knowledge, including how to define operators and inference rules. Section 4 describesPRODIGY’s control structure in detail, showing exactly how the search proceeds.Section 5 explains how to use control rules to guide the search, and how to write them. To run the example from Section 2, skip the sections mentioned above and go directly to Section 6. Sections 6.2 and 6.5 describe how to load and run the system, and how to use the trace produced by it. Section 7 gives some hints on how 。