Acetyl-modulated architecture for evolutionary robotics

Abstract

Control modularity is an important tool for improving the organisation and potential task resolution in Evolutionary Robotics. Neuro-controllers are divided into many sections or many coordinated networks to solve a particular task. In this paper, we present a model for control evolutionary robots inspired by the effects of acetylcholine neurotransmitter, chemical synapse, renshaw cells, and based on artificial neural networks. The performance of our model is compared to the other two implementations: a) a system controlled by one single neural network and b) a control system based on one neural network divided into different sections. A garbage collection and room swap tasks were used for the experimentation in order to validate the proposed model. As for the simulator, we decided to use a commercial platform as Webots, which includes a virtual e-puck robot in a 3D environment.