Resumen de Open-Ended Learning of Reactive Knowledge in Cognitive Robotics Based on Neuroevolution

Alejandro Romero Montero, Francisco Bellas Bouza , Richard J. Duro Fernández

• Reactive knowledge corresponds to implicit knowledge in the human brain, that is, unconscious knowledge such as reflexes that are executed without “thinking”. It is a key aspect in human development, and it is also a key aspect in cognitive architectures for robots, mainly because it avoids inefficient action selection procedures and allows addressing higher-level cognitive processes that make use of it. This paper deals with the acquisition of this type of knowledge in a cognitive architecture for open-ended learning. We propose a method for the learning of policies (reactive knowledge) trough evolution from deliberative models by means of neuroevolution. It is interesting to see in the results presented that this approach of learning reactive knowledge instead of exhaustively selecting the appropriate action every instant of time provides equivalent/better results and more efficient action sequences.