Flexible Decision-Making for Autonomous Agent Through Computation of Urgency in Time-Critical Domains

실시간 환경에서 긴급한 정도의 계산을 통한 자율적인 에이전트의 유연한 의사결정

  • 노상욱 (가톨릭대학교 컴퓨터정보공학부)
  • Published : 2004.09.01

Abstract

Autonomous agents need considerable computational resources to perform rational decision-making. The complexity of decision-making becomes prohibitive when large number of agents are present and when decisions have to be made under time pressure. One of approaches in time-critical domains is to respond to an observed condition with a predefined action. Although such a system may be able to react very quickly to environmental conditions, predefined plans are of less value if a situation changes and re-planning is needed. In this paper we investigate strategies intended to tame the computational burden by using off-line computation in conjunction with on-line reasoning. We use performance profiles computed off-line and the notion of urgency (i.e., the value of time) computed on-line to choose the amount of information to be included during on-line deliberation. This method can adjust to various levels of real-time demands, but incurs some overhead associated with iterative deepening. We test our framework with experiments in a simulated anti-air defense domain. The experiments show that the off-line performance profiles and the on-line computation of urgency are effective in time-critical situations.

자율적인 에이전트들은 이성적인 의사결정을 위하여 상당한 양의 계산자원을 필요로 하며, 실시간 환경에서 항상 최적의 행동을 수행하는 완벽하게 이성적인 에이전트(rational agent)의 구현은 실질적으로 가능하지 않다. 이러한 실시간 문제 해결기법에서의 전통적인 접근 방식은 미리 정의된 규약에 의존한 조건-행동 추론 방식이다. 조건-행동 추론 방식은 주어진 상황에 빠르게 반응하지만, 문제 영역이 다양하거나 문제의 재설계가 필요한 경우에는 아무런 해법을 갖지 못한다. 따라서 이러한 문제점을 해결하기 위해 본 논문에서는 주어진 행동들의 유틸리티를 실시간에 계산하고, 긴급한 정도(urgency)를 측정하여 상황이 긴급할 경우에는 더 이상의 계산을 중단하고 즉각적인 행동을 취하며, 반면에 상황이 긴급하지 않을 경우에는 최선의 의사결정을 위하여 추가적인 정보를 고려하여 더 바람직한 행동을 결정하는 방법론을 제안한다. 제안한 방법론의 평가를 위하여 시간 제약적인 환경에서 최선의 의사결정을 수행하는 실질적이며 유연한 에이전트를 구현하고자 한다.

Keywords

References

  1. Agre, P.E. and Chapman, D., Pengi: An implementation of a theory of activity, In Proceedings of the National Conference on Artificial Intelligence, Seattle, Washington, pp. 268-272, 1987
  2. Fox, J. and Krause, P., Symbolic decision theory and autonomous systems, In Proceedings of the 7th Conference on Uncertainty in Artificial Intelligence, UCLA, California, pp. 103-110, July 1991
  3. Russell, S. J. and Wefald, E. H., Principles of Metareasoning, Journal of AI, Volume 49, pp, 361-395, 1991 https://doi.org/10.1016/0004-3702(91)90015-C
  4. Russell, S.J. and Subramanian, D., Provably bounded-optimal agents, Journal of Artificial Intelligence Research, Volume 2, pp, 575-609, 1995
  5. Russell, S.J., Rationality and Intelligence, Artificial Intelligence, Volume 94, pp. 57-77, 1997 https://doi.org/10.1016/S0004-3702(97)00026-X
  6. Rao, A.S. and Georgeff, M.P., BDI agents: From theory to practice, In Proceedings of the 1 st International Conference on Multiagent Systems, pp. 312-319, July 1995
  7. Dean, T., Decision-theoretic planning and markov decision processes, Technical Report, Brown University, Computer Science Department, Providence, RI, 1994
  8. Zilberstein, S. and Russell, S.J., Optimal composition of real-time systems, Artificial Intelligence, vol. 82, no. 1, pp. 181-213, 1996 https://doi.org/10.1016/0004-3702(94)00074-3
  9. Noh, S. and Gmytrasiewicz, P.J., Uncertain Knowledge Representation and Communicative Behavior in Coordinated Defense, Lecture Notes in Artificial Intelligence 1916, Issues In Agent Communication. pp. 281-300, Springer, 2000
  10. Macfadzean, R.H.M., Surface-Based Air Defense System Analysis, Artech House, 1992
  11. Gmytrasiewicz, P.J. and Durfee, E.H., Rational coordination in multi-agent environments, Autonomous Agents and Multiagent Systems Journal, vol. 3, pp. 319-350, 2000 https://doi.org/10.1023/A:1010028119149
  12. Gmytrasiewicz, P.J., Noh, S., and Kellogg, T., Bayesian update of recursive agent models, User Modeling and User-Adapted Interaction: An International Journal, vol. 8, no. 1/2, pp. 49-69, 1998 https://doi.org/10.1023/A:1008269427670