A layered approach to learning intelligent behaviours in rescue robot simulation system using fuzzy logic and neural networks

RoboCup Rescue Simulation System is a particularly challenging domain for studying multi agent system and multi agent learning. Machine learning has become a key solution to complicated multi agent tasks. In this paper, using machine learning as a tool for arriving at intelligent and efficient behaviors for Rescue robots involves layering increasingly complex learning behaviors. We describe multiple levels of learned behaviors. First the robots try to lean basic knowledge about their environment's characteristics like the spreading speed of tire in the city after earthquake, or their ability to extinguish fires in different situations. ANN has been used to achieve these goals. Afterwards, using these learned components, they learn low level skills for lire extinguishment. Finally, in the next level they exploit fuzzy logic for planning their high level strategy toward their goal.