Interactive Object Recognition Using Proprioceptive Feedback

This paper proposes a method for interactive recog- nition of household objects by a robot using proprioceptive feedback. In our experiments, the robot observed the changes in its proprioceptive stream while performing five exploratory behaviors (lift, shake, drop, crush, and push) on 50 common household objects (e.g., bottles, cans, balls, toys, etc.). Spe cifically, the robot used its own joint torques recorded during each interaction to recognize the object that it was manipulating. The results show that the robot can learn to recognize objects solely from the proprioceptive information obtained while interacting with them. Furthermore, by applying multiple behaviors on the same object, the robot was able to significantly improve its object recognition accuracy. Overall, the results show that proprioception should be considered as an important source of information for object recognition and object manipulation tasks.