Learning Perceptual Organization with a Developmental Robot

This paper presents a biologically-inspired model of self-organization for robotic intermediary vision. Two mechanisms are under concern. First, the development of low-level local feature detectors adapted to the sensory input signal statistics in order to perform a piecewise categorization of the sensory signal. Second, the hierarchical grouping of these local features in a holistic perception. While the grouping mechanism is expressed as a classical agglomerative clustering, underlying similarity measures are not pre-given but developed from the signal statistics. Segmentation results are therefore adapted to the robot\´s experience. Based on information-theory, a stopping criterion that expresses a "good" abstraction level is proposed. Proposed mechanisms are illustrated with examples in both color and edge segmentation.