Non-contact target acquisition and object identification for robotic grasping

A CCD stereo camera is positioned so that it can monitor the work space of a robot. The camera can locate objects within the work space with sufficient accuracy so that an object's major dimensions, as well as its spatial coordinates relative to the camera can be determined. A robot can then grasp the selected object with its end effector. An object recognition program detects automatically the presence of target objects for grasping. After the desired object has been identified, its location relative to the robot gripper and its size are determined. The stereo camera's disparity image provides a depth map of the scene (including the spatial coordinates of the target object). The spatial distance vector is computed and an appropriate command sequence for the robot motion is transmitted to the low-level robot controller. The supervisory program is implemented in C-language on a PC, with program modules handling data acquisition, image analysis, and bidirectional communication with the robot. Experimental results show that the end effector positioning error is on the same order of magnitude as the positioning inaccuracy of the PUMA 560 robot used for the experiments. The paper describes design criteria, software structure and organization, the decision and rule structure of the control system, and experimental results.