Multilevel Darwinist Brain and Autonomously Learning to Walk

Abstract

This paper is devoted to the application of the Multilevel Darwinist Brain to the autonomous learning and generation of gait patterns by a hexapod robot. This robot must learn to walk and reach an objective without any pre-programmed fitness function or behaviour. It develops a model of its world and of itself by interacting with the world, very much like a child, and this allows it to choose the appropriate actions to fulfil its motivations. As this interaction progresses its actions in the world become more focused and efficient and after some time it is able to generate the optimal gait and control the swing amplitude of the legs so as to reach its objective every time, independently of where it is located with respect to its body. To achieve these behaviours we have developed a Darwinist cognition mechanism operating over artificial neural networks that allows the robot to learn from its interaction with the world generating original solutions and making use of experience.