The Emerge of Disentanglement in the Conceptual Space during Sensorimotor Interaction

The conceptualization of information obtained from the external world is the foundation of language development for intelligent agents. The basic target of this process is to summarize the common properties of the environment and to further name it to describe those properties in the future. To realize this purpose, a new learning model is introduced for the disentanglement of several sensorimotor concepts (e.g. sizes, colours, shapes of objects) while the causal relationship is being learnt during interaction without much prior experience and external instructions. This learning model links predictive deep neural models and thevariationalauto-encoder(VAE)andprovidesthepossibilityabouttheindependentconceptscanbeextractedanddisentangled from both perception and action. And such extraction is further learnt by VAE to memorize their common statistical features. We examine this model in the affordance learning setting, where the robot is trying to learn to disentangle the concepts about shapes of the tools and objects. The results show that such concepts can be found in the neural activities of the β-VAE unit, which indicates that using similar VAE models is a promising way to learn the concepts and thereby to learn the causal relationship of the sensorimotor interaction.