Learning to Perceive the World as Articulated: An Approach for Hierarchical Learning in Sensory-Motor Systems

This paper describes how the internal representation of the world can be self-organized in modular and hierarchical ways in a neural network architecture for sensory-motor systems. We develop an on-line learning scheme – the so-called mixture of recurrent neural net (RNN) experts – in which a set of RNN modules becomes self-organized as experts in multiple levels in order to account for the different categories of sensory-motor flow which the robot experiences. The proposed scheme was examined through simulation experiments involving the navigatiois learning problem, in which a robot equipped with range sensors traveled around rooms of different shape. It was shown that representative building blocks or “concepts” corresponding to turning right and left at corners, going straight along corridors and encountering junctions are self-organized in their respective modules in the lower level network. In the higher level network, the “concepts” corresponding to traveling in different rooms are self-organized by combining the “concepts” obtained in the lower level into sequences. The robot succeeded in learning to perceive the world as articulated at multiple levels through its recursive interactions.