Robots use Object Map


Results achieved with “Dense Object Nets” wrt some rudimentary form of eye-hand coordination. It is supposed to be object orientation invariant and can map a features to unseen objects in the class.
The article includes an impressive video showing the power of the mapping.

Robot picks up stuff


That is amazing! We are definitely on the right track. HTMs are going to be able to do this type of task very efficiently, and should be able to generalize and interact with new objects never seen before by relating them to already learned objects.