We have designed and implementedareal-time binocular tracking system which uses two independent cues commonly found in the primary functions of biological visual systems to robustly track moving targets in complex environments, without a-priori knowledge of the target shape or texture: a fast optical flow segmentation algorithm quickly locates independently moving objects for target acquisition and provides a reliable velocity estimate for smooth tracking. In parallel, target position is generatedfrom the output of a zero-disparity filter where a phase-based disparity estimation technique allows dynamic control of the camera vergence to adapt the horopter geometry to the target location. The system takes advantage of the optical properties of our custom-designed foveated wide-angle lenses, which exhibit a wide field of view along with a high resolution fovea. Methods to cope with the distortions introduced by the space-variant resolution, and a robust real-time implementation on a high performance active vision head arepresented. 1

Accurate oculomotor control is one of the essential pre-requisites of successful visuomotor coordination. Given the variable nonlinearities of the geometry of binocular vision as well as the possible nonlinearities of the oculomotor plant, it is desirable to accomplish accurate oculomotor control through learning approaches. In this paper, we investigate learning control for a biomimetic active vision system mounted on a humanoid robot. By combining a biologically inspired cerebellar learning scheme with a state-of-the-art statistical learning network, our robot system is able to acquire high performance visual stabilization re exes after about 40 seconds of learning despite signi cant nonlinearities and processing delays in the system. 1

This paper proposes a neural architecture for a robot in order to learn how to imitate a sequence of movements performed by another robot or by a human. The main idea is that the imitation process does not need to be given to the system but can emerge from a mis-interpretation of the perceived situation at the level of a simple sensory-motor system. The robot controller is based on a PerAc (Perception-Action) architecture. This architecture allows an autonomous robot to learn by itself sensory-motor associations with a delayed reward. Here, we show how the same architecture can also be used by a "student" robot to learn to imitate another robot allowing the student robot to discover by itself solutions to a particular problem or to learn from another robot what to do. We discuss the difficulty linked to the segmentation of the actions to imitate. This imitation problem is demonstrated by a task of learning a sequence of movements and their precise timing. Another interesting aspect of this work is that the neural network (N. N.) used for sequence learning is directly inspired from a brain structure named the hippocampus and mainly involved in memory processes (Banquet et al. 1997). We discuss the importance of imitation processes for the understanding of our high level cognitive abilities linked to self-recognition and to the recognition of the other as something similar to me. 1

une s'equence motrice et de son timing pr'ecis. Un autre int'eret de ce travail est que le r'eseau de neurones utilis'e pour l'apprentissage des s'equences est directement inspir'e d'une structure c'er'ebrale, l'hippocampe, principalement impliqu'ee dans les processus de m'emoire. Nous discutons l'importance des processus d'imitation pour la compr 'ehension de nos capacit'es cognitives de "haut niveau" li'ees `a la reconnaissance de soi et `a la reconnaissance des autres comme quelque chose de similaire `a soi. Mots cl'es: imitation, apprentissage de s'equences temporelles, neuro-cybern'etique 1 Introduction Depuis plus d'une quarantaine d'ann'ees maintenant, les chercheurs en Intelligence Artificielle (IA) et en robotique essaient de cr'eer des syst`emes capables d'apprendre par eux meme. Le bilan de ces travaux est tr`es contrast'e. Un grand nombre d'algorithmes d'optimisation ont vu le jour. Dans la plupart des cas, en robotique, apprendre `a r'esoudre un probl`e

www.taros.org.uk