A neuromorphic control module for real-time vergence eye movements on the iCub robot head

Abstract

We implemented a cortical model of vergence eye movements on a humanoid robot head (iCub). The proposed control strategy resorts on a computational substrate of modeled VI complex cells that provides a distributed representation of binocular disparity information. The model includes a normalization stage that allows for a vergence control independent of the texture of the object and of luminance changes. The disparity information is exploited to provide a signal able to nullify the binocular disparity in a foveal region. Experimental tests demonstrated that the robot head executes effective vergence movements, on a visual stimulus posed within the peripersonal space, that falls in the perifoveal field of view. The algorithm works in real-time, and provides a stable control, robust with respect to the mechanical inaccuracies that affect the mechanical system. The eye posture resulting from the vergence movement, is instrumental for an active vision system to optimally perceive the 3D structure of the environment from disparity. The similarity of the vergence speed profile and trajectories of the fixation point with respect to psychophysical data, allows us to use the system for a validation of the neurophysiological models for triggering vergence eye movements.