Generating Macro Operators for Decision-Theoretic Planning

ion in probabilistic domains has long been rec-ognizedas an effective tool to faciliate faster reasoning.In this paper we focused on the technical side of ob-taining a correct macro abstraction; a comprehensivetheory of abstracting probabilistic actions is presentedin [2]. Empirical results of applying the technique toseveral domains showed substantial reduction in thecomplexity of planning [2]. The sequential abstrac-tion method presented here could be easily adopted touse in both open-loop probabilistic planning [6] andMarkov models [9]. We have ignored the question ofabstracting schematic actions. Further work will befocused on this problem, as well as on investigatingabstractions based on qualitative properties of actions.References [2][1] L. Chrisman. Abstract probabilistic modelingof action. In Proceedings of the First Interna-tional Conference on Artificial Intelligence Plan-ning Systems, pages 28-36, June 1992.A.H. Doan and P. Haddawy. Decision-theoreticrefinement planning: Principles and applica-tion. Technical Report TR-95-01-01, Dept. ofElect. Eng. & Computer Science, University ofWisconsin-Milwaukee, January 1995. Availablevia anonymous FTP from pub/tech_reports atftp.cs.uwm.edu. [3] R.E. Fikes, P.E. Hart, and N.J. Nilsson. Learningand executing generalized robot plans. ArtificialIntelligence, 3(4):251-288, 1972.[4] P. Haddawy and A.H. Doan. Abstracting proba-bilistic actions. In Proceedings of the Tenth Con-ference on Uncertainty in Artificial Intelligence,pages 270-277, Seattle, July 1994. Availablevia anonymous FTP from pub/tech_reports atftp.ca.uwm.edu. [5] C.E. Kahn and P. Haddawy. Optimizing diag-nostic and therapeutic strategies using decision-theoretic planning: Principles and applications.In Proceedings of the Eighth World Congresson Medical Informatics (Medinfo ’95), Vancouver,July 1995. (to appear). [6] N. Kushmerick, S. Hanks, and D. Weld. An al-gorithm for probabilistic least-commitment plan-ning. In Proceedings of the Twelfth NationalConference on Artificial Intelligence, pages 1073-1078, Seattle, 1994.[7] S. Minton. Selectively generalizing plans for prob-lem solving. In IJCAI85, pages 596-599, 1985.[8] J.M. Siskind and D.A. McAllester. Nondetermin-istic lisp as a substrate for constraint logic pro-gramming. In Proceedings of the Eleventh Na-tional Conferenceon Artificial Intelligence, pages133-138, Washington,DC, July 1993.[9] David E. Smith and Mike Williamson. Repre-sentation and evaluation of plans with loops. InAAAI Spring Symposium 95Extending Theoriesof Actions, Stanford, CA, 1995.[10] S. Thiebaux, J. Hertzberg, W. Shoaff, andM. Schneider. A stochastic model of actions andplans for anytime planning under uncertainty. In-ternational Journal of Intelligent Systems, 1994.To appear.