Genetic Algorithm Integrated with Neural-Network for Tolman Mouse Robot Navigation

Abstract

Mapping building plays an important role in robot navigation, in order to facility agent with high level intelligence and mimic the major function of rat’s space cell, a new mechanism of Genetic Algorithm Integrated with Neuralnetwork(GAIN) was proposed in this paper. Considering the scenario of unknown environment when agent explores the map, weights in neural network remain the same during agent’s lifecycle and will be optimized by genetic algorithm. Several simulation was performed on Unity platform, especial the Tolman mouse maze experiment, including cross position learning experiment, spatial orientation experiment and roundabout experiment, had been reproduced by agent other than real rat. Furthermore, some extension of experiment has also done to further prove the feasibility of the algorithm. Simulation results verified the proposed GAIN algorithm can endow agent with cognitive map function.