Extraction of multi-modal object representations in a robot vision system

We introduce one module in a cognitive system that learns the shape of objects by active exploration. More specifically, we propose a feature tracking scheme that makes use of the knowledge of a robotic arm motion to: 1) segment the object currently grasped by the robotic arm from the rest of the visible scene, and 2) learn a representation of the 3D shape without any prior knowledge of the object. The 3D representation is generated by stereo-reconstruction of local multi-modal edge features. The segmentation between features belonging to the object those describing the rest of the scene is achieved using Bayesian inference. We then show the shape model extracted by this system from various objects.