Thinking With Your Body: Modelling Spatial Biases in Categorization Using a Real Humanoid Robot

Abstract

This paper presents a model of sensorimotor learning grounded in the sensory streams of a real humanoid robot (the iCub robot). The robot participates in a replication of two developmental psychology experiments, in which it is shown how spatial cues are sufficient for associating linguistic labels with objects. The robot, using auto-associated self-organizing maps connecting is perceptual input and motor control, produces similar performance and results to human participants. This model confirms the validity of a body centric account of the linking of words to objects as sufficient to account for the spatial biases in learning that these experiments expose.