Abstract— The purpose of this work is to investigate the applicability of a visual tracking model on humanoid robots in order to achieve a human-like predictive behavior. In humans, in case of moving targets the oculomotor system uses a combination of the smooth pursuit eye movement and saccadic movements, namely “catch up” saccades to fixate the object of interest This work aims to validate the “catch up” saccade model in order to obtain a human-like tracking system able to correctly switch from a zero-lag predictive smooth pursuit to a fast orienting saccade for the position error compensation. Experimental results on the iCub simulator show several correspondences with the human behavior. INTRODUCTION he primate visual system is characterized by binocular visual fields and a space-variant resolution retina with a high-resolution fovea that offers considerable advantages for a detailed analysis of visual objects, together with effective visuo-motor control [1]. The space-variant resolution of the retina requires efficient eye movements for correct vision. Two forms of eye movements — saccades and smooth pursuit — enable us to fixate the object on the fovea. Saccades are high-velocity gaze shifts that bring the image of an object of interest onto the fovea. Smooth pursuit occurs when the eyes track a moving target with a continuous motion, in order to minimize the image slip in the retina and make it perceptually stable. Smooth pursuit movements cannot normally be generated without a moving stimulus although they can start a short moment before the target is expected to appear [2]. Smooth pursuit is complicated by the fact that the initial visual processing in the human brain delays the stimulus by approximately 100 ms before it reaches the visual cortex [2][3]. In primates, with a constant velocity or a sinusoidal target motion, the smooth pursuit gain, i.e. the ratio of tracking velocity to target velocity, is almost 1.0 [4]. This cannot be achieved by a simple visual negative feedback controller due to the long delays (around 100 ms in the human brain), most of which are caused by visual information processing. In the monkey brain, the neural pathway that mediates smooth-pursuit eye movements, described in [5], starts in the primary visual cortex (V1) and extends to the middle temporal area (MT) that serves as generic visual motion processor. It contributes to smooth pursuit measuring the Corresponding author: Egidio Falotico Scuola Superiore Sant’Anna e.falotico@sssup.it target motion in retinal coordinates[6][7][8]. By contrast, the middle superior temporal area (MST) seems to contain the explicit representation of object motion in world centred coordinates [9]. Recent works [10] demonstrate that this area is responsible for target dynamics prediction. Cortical eye fields are also involved in smooth pursuit [11]; in particular the frontal eye field (FEF) can modulate the gain control [12][13][14] that determines how strongly pursuit will respond to a given motion stimulus. Saccades are fast eye movements (maximum eye velocity > 1000 deg/sec) that allow primates to shift the orientation of the gaze using the position error (difference between the target position and the eye position) [15]. The duration of the saccadic movement is very short (30-80ms), so they cannot be executed with continuous visual feedback. The saccade generation consists in a sensorimotor transformation from visual space input to the motor command space. That transformation involves many brain areas from the superior colliculus (SC) to the cerebellum. Some of these areas are similar to those involved in the smooth pursuit generation [16]. Usually smooth pursuit is executed for predictable target motion rather the saccades are used in correspondence of static target. In case of moving target the oculomotor system uses a combination of the smooth pursuit eye movement and saccadic movement, namely “catch up” saccades to fixate the object of interest. Recent studies investigate the mechanisms underlying the programming and the execution of catch-up saccades in humans [17][18][19]. This work proposes an implementation on iCub robot of a model for these movements. From the robotic point of view there are different implementations of these eye movements [20][21][22][23][24].This work aims to validate the “catch up” saccade model in order to obtain a human-like tracking system able to correctly switch from a zero-lag predictive smooth pursuit to a fast orienting saccade for the position error compensation. VISUAL TRACKING MODEL This work proposes the integration of different systems in order to obtain a human like behavior of a predictive smooth pursuit of a dynamic target. The purpose of this work is to investigate the applicability of the visual tracking model on humanoid robots in order to achieve a human-like predictive behavior that can adapt itself to changing of environment and to learn from the experience. This model is able to predict target trajectories that present a second order dynamics. It is possible to extend this model to cope with Implementation of a bio-inspired visual tracking model on the iCub robot Egidio Falotico, Davide Zambrano, Giovanni Muscolo, Laura Marazzato, Paolo Dario, Cecilia Laschi T CONFIDENTIAL. Limited circulation. For review only. Preprint submitted to 19th IEEE International Symposium on Robot and Human Interactive Communication. Received April 1, 2010.