Interactive Perception for Cluttered Environments

Abstract

Robotics research tends to focus upon either non-contact sensing or machine manipulation, but not both. This paper explores the benefits of combining the two by addressing the problem of extracting and classifying unknown objects within a cluttered environment, such as found in recycling and service robot applications. In the proposed approach, a pile of objects lies on a flat background, and the goal of the robot is to sift through the pile and classify each object so that it can be studied further. One object should be removed at a time with minimal disturbance to the other objects. We propose an algorithm, based upon graph-based segmentation and stereo matching, that automatically computes a desired grasp point that enables the objects to be removed one at a time. The algorithm then isolates each object to be classified by color, shape and flexibility. Experiments on a number of different objects demonstrate the ability of classifying each item through interaction and labeling them for further use and study.