Learning Bipedal Walking Through Morphological Development

Morphological development has shown its efficiency in improving learning and adaptation to the environment in natural organisms from infancy to adulthood. In the case of robot learning, this is not so clear. The results of a series of experiments that have been carried out in previous work have allowed us to extract, from an analytical perspective, some notions about how and under what conditions morphological development may influence learning. In this paper, we want to adopt an engineering or synthesis perspective and test whether these notions can be used to construct a successful morphological development strategy for a difficult task: learning bipedal locomotion. In particular, we have addressed learning to walk in a 14 degrees of freedom NAO type robot and have designed a morphological development strategy to this end. The results obtained have allowed us to validate the relevance of the assumptions made for the design and implementation of a morphological development strategy.