Representing Affordances of Objects Through Active Touch

An intelligent robot must build an understanding of its environment with minimal human mediation. We investigate the characterization of objects in terms of their affordances�the ways in which an agent can functionally interact with objects. To effect this goal, we represent objects as expectations with respect to sensorimotor schema�collections of data that describe how an agent expects its interaction with the world to affect its perception of the world. We are constructing novel robotic manipulators that have enhanced abilities to actively touch their environment to develop these ideas. This work will contribute to our larger effort of grounding natural language in the physical environment.