How can a robot learn to perceive an unknown sensorimotor world, and learn to act effectively in it? How can it start with pixel-level perception and motor babbling, and create its own higher-level concepts of objects and actions applied to them, and places and motions connecting them? What primitive capabilities should the individual learner possess, and how could these be learned over evolutionary time?
What are suitable hierarchical representations for the 3D structure of object instances or classes that can be exploited for recognition and action affordances?
How can different kinds of spatial constraints be used for joint recognition of scenes and the objects within them? We invite applications for a post-doctoral position at the University of Michigan to address these and related problems, working with two faculty rese ...Read the full article