Cloth Black Acrylic Cutting BoardBubble Wrap Plastic Case Plastic Dispenser

To perform useful tasks in everyday human environments, rob ots must be able to both understand and communicate the sensa tions they experience during haptic interactions with objects. T oward this goal, we augmented the Willow Garage PR2 robot wit h a pair of SynTouch BioTac sensors to capture rich tactile signals d uring the execution of four exploratory procedures on 60 hou sehold objects. In a parallel experiment, human subjects blindly t ouched the same objects and selected binary haptic adjectiv es from a predetermined set of 25 labels. We developed several machin e-learning algorithms to discover the meaning of each adjec tive from the robot’s sensory data. The most successful algorithms we re those that intelligently combine static and dynamic comp nents of the data recorded during all four exploratory procedures. T he best of our approaches produced an average adjective clas sifi ation F1 score of 0.77, a score higher than that of an average human sub ject.