Neuroevolutionary Motivational Engine for Autonomous Robots

Abstract

In this paper we present a motivational engine based on neuroevolution that permits a robot to autonomously obtain the appropriate value functions for its interaction with an environment. This motivational engine permits the combination of exploratory and exploitative drives through a certainty based approach. It allows for the exploration of the state space in order to determine where rewards can be obtained and provides an evolutionary based mechanism for constructing the corresponding value functions. An example considering a real robot in a real environment that implies the construction of several value functions is analyzed.