Dynamic Autonomous Agent Positioning Based on Computational Intelligence


In this paper, we present a modular neural network for learning formation strategy in multi-agent systems. A supervised learning method is devised to train the modular neural network in order for a group of agents to learn formation strategy in an environment. At first, the environment conditions are separated into some different parts called contexts in this paper. Consequently, each agent employs a neural network to learn the sequence of actions of expert, according to the present context. After the training process, agents would be able to imitate human behavior in similar conditions. As a result, an intelligent model of human behavior is extracted which contributes in building autonomous agents. This framework increases the robustness and efficiency of the multi-agent system while providing the system with redundancy, reconfiguration ability and structure flexibility. The fuzzy ARTMAP neural network combines a unique set of computational abilities that are needed to function autonomously in a changing world. These Characters lead us to use this network in learning process. Therefore, the modular fuzzy ARTMAP neural network is used to extract expert knowledge in formation strategy. In particular, the proposed framework is applied to soccer robots and its generalization capability is evaluated with datasets from which several data points are randomly removed.

6 Figures and Tables

Cite this paper

@inproceedings{Korjani2001DynamicAA, title={Dynamic Autonomous Agent Positioning Based on Computational Intelligence}, author={Mohammad Mehdi Korjani and Ahmad Afshar and Mehrab Norouzitallab and Mohammad Bagher Menhaj}, year={2001} }