A Study of Autonomous Agent Navigation Algorithms in the ANIMAT Environment
This paper deals with the examination of possibilities of the algorithm of autonomous agent navigation. The ANIMAT problem was developed to study the possibilities of ZCS classifier system learning. A ZCS classifier system is purely reactive. For this system it is essential to have a possibility to identify the global state of the environment by the input message. This reduces the application area of the system and its ability to solve complex tasks. The present paper proposes to build an alternative learning algorithm that would be able to cope with the above mentioned problems. The effectiveness of the suggested algorithm is tested via practical experiments. In the experiments performed the algorithm has demonstrated its essential superiority over a classifier system with temporary memory.
KeywordsOptimal Policy Target State Classifier System Belief State Agent Learning
Unable to display preview. Download preview PDF.
- 1.Holland, J. H. Adaptation. In R. Rosen & F.M. Snell (Eds.), Progress in theoretical biology, 4. New York: Plenum, 1976Google Scholar
- 2.Wilson, S. W. Knowledge growth in an artificial animal. Proceeding of the First International Conference on Genetic Algorithms and Their Applications (pp. 16–23). Hillsdale, New Jersey: Lawrence Erlbaum Associates, 1985Google Scholar
- 6.Lanzi, P. L. Adding Memory to XCS. In: Proceedings of the IEEE Conference on Evolutionary Computation. IEEE Press, 1998Google Scholar
- 7.Wilson, S. W. The Animat Path to AI. In From Animals to Animats: Proceedings of The First International Conference on Simulation of AdaptiveBehavior (pp. 15–21), J.-A. Meyer and S.W. Wilson, eds., Cambrige, MA: The MIT Press, Bradford Books,1991Google Scholar
- 9.Cassandra, A. R. Optimal policies for partially observable Markov decision processes. Technical Report CS-94–14, Brown University, Department of Computer Science, Providence RI, 1994Google Scholar
- 10.Cassandra, A. R. & Kaelbling, L. P. & Littman, M. L. Optimal policies for partially observable stochastic domains. In Proceedings of the Twelfth National Conference on Artificial IntelligenceSeattle, WA, 1994Google Scholar
- 11.Pchelkin, A. A study of autonomous agent navigation possibilities in the Animat environment. Master’s Thesis, The University of Latvia, supervised by Prof. Arkady Borisov, 2001Google Scholar
- 12.Balcázar J. L. & Díaz, J. & Galvaldà, R. & Watanabe, O. Algorithms for Learning Finite Automata from Queries: A Unified ViewTechnical Report TR96–0017 Department of Computer Science, Tokyo Institute of Technology, 1996Google Scholar