Martha E. Pollack

Research on planning within AI has led to the development of computational techniques for generating a plan to achieve a specified goal from a specified initial state, given definitions of the available actions. Until recently, however, much of this research has been governed by a number of simplifying assumptions, notably that there is no indeterminacy either in the planning agent's knowledge, in the effects of its actions, or in the notion of goal satisfaction, and that are no changes either in the goals presented to the agent or in its environment. This project has been investigating strategies that enable planning agents to cope with environments in which these assumptions do not hold. The strategies we have studied respect, and in many cases even exploit, the fact that all agents have computational resource limits: they cannot perform arbitrarily large computations in a fixed, finite amount of time.

The problem of resource-limited planning for dynamic, unpredictable environments in a large one, and we have investigated and made progress on a number of different aspects of it.

• Development and analysis of meta-level control strategies for planning in single-agent, dynamic environments. Some references include:
  • Y. Ronen, D. Mosse', and M. E. Pollack, Value-Density Algorithms for the Deliberation Scheduling Problem SIGART Bulletin, 7(2), 1996.

  • M. E. Pollack, D. Joslin, A. Nunes, S. Ur, & E. Ephrati, Experimental Investigation of an Agent Commitment Strategy, Univ. of Pittsburgh Tech. Report Dept. of Computer Science 94-31, June 1994.

  • M. E. Pollack, The Uses of Plans , Artificial Intelligence, 57(1):43-68, 1992.

• Generalization of those strategies for use in multi-agent, dynamic environments. Some references include:
  • C. Bicchieri, M. E. Pollack, C. Rovelli, and I. Tsamardinos, "The Potential for the Evolution of Cooperation among Web Agents" To appear in International Journal of Human-Computer Studies, 1997.

  • E. Ephrati, M. E. Pollack, and S. Ur, Deriving Multi-Agent Coordination through Filtering Strategies Proceedings of the 14th International Joint Conference on Artificial Intelligence , August 1995.

  • E. Ephrati, M. E. Pollack, and J. S. Rosenschein, A Tractable Heuristic that Maximizes Global Utility through Plan Combination The First International Conference on Multi-Agent Systems , June 1995.

• Development, implementation, and analysis of testbed platforms for experimentally studying dynamic planning. Some references include:
  • M. E. Pollack, Planning in Dynamic Environments: The DIPART System In A. Tate, editor, "Advanced Planning Technology" AAAI Press, 1996.

  • S. Hanks, M. E. Pollack and P. Cohen, Benchmarks, Testbeds, Controlled Experimentation, and the Design of Agent Architectures , AI Magazine , 14(4):17-42, 1993.

• Design of strategies for search-control in planning, and extensive experimental comparison of competing search-control strategies. Some references include:
  • M. E. Pollack, D. Joslin, and M. Paolucci, "Flaw Selection Strategies for Partial-Order Planning"Univ. of Pittsburgh Dept. of Computer Science Technical Report 96-20R, Feb. 1997. (Also under review for publication.)

  • E. Ephrati, M. E. Pollack, and M. Milshtein A Cost Directed Planner: Preliminary Report In Proceedings of the 13th National Conference on Artificial Intelligence, August 1996.

  • D. Joslin and M. E. Pollack, Least-Cost Flaw Repair AAAI-94 , Seattle, Aug., 1994.

• Construction of new, constraint-based methods for plan generation. Some references include:
  • D. Joslin and M. E. Pollack, Is "Early Commitment" in Plan Generation Ever a Good Idea? In Proceedings of the 13th National Conference on Artificial Intelligence, August 1996.

  • D. Joslin and M. E. Pollack, Active and Passive Postponement of Decisions in Plan Generation Proceedings of the 3rd European Workshop on Planning , September 1995.

• Design and analysis of strategies for selecting critical contingencies in conditional planning. Some references include:
  • N. Onder and M. E. Pollack, Contingency Selection in Plan Generation" Proceedings of the AAAI Fall Symposium on Plan Execution," Nov. 1996.

  pollack@cs.pitt.edu
  
  
  

  ---

  
  Return to ITO Workshop Abstracts
  
  Return to ITO Workshop Home Page