Distance Perception in Mobile Robots as an Emergent Consequence of Visuo-Motor Cycles Using Forward Models

Abstract

Distance perception has been studied for long in Cognitive Sciences and stills a complex problem. According to some research hypothesis in psychology, perception of distance is not a geometrical process but emerges as an association of multimodal sensory information. Instead of giving the robot the explicit means to model the external metric or topology of the free space or to exploit geometrical information from stereo vision techniques, we aim at validating internal models that code for spatial relationships accounting for a distance perception capability in the robot, by means of the prediction and re-enaction of visuo-motor cycles as the fundamental cognitive process. We use foveated images as the sensory situation together with current motor commands as input to a forward model that predicts the next images and a proximal tactile sensory situation, thus coding a notion of distance in robot motor coordinates. Experimental results are encouraging in our quest to modeling distance perception in mobile robots.