A robot can struggle to discover objects in its surroundings when it relies on computer vision alone. But by taking advantage of all of the information available to it — an object’s location, size, shape and even whether it can be lifted — a robot can continually discover and refine its understanding of objects, say researchers.