Simulation of Rat Behavior in a Light-Dark Box via Neuroevolution

Abstract

The light-dark box is a widely used test for the investigation of animal behavior commonly used to identify and study anxious-like behavioral patterns in rodents. We propose a neuroevolution model for virtual rats in a simulated light-dark box. The virtual rat is controlled by an artificial neural network (ANN) optimized by a genetic algorithm (GA). The fitness function is given by a weighed sum of two terms (punishment and reward). By changing the weight of the punishment term, we are able to simulate the effects of anxiolytic/anxiogenic drugs on rats. We also propose using GAs to optimize the number of the ANN hidden neurons and sensors for the virtual rat. According to the experiments, the best results are obtained by ANNs combining both luminosity and wall sensors.