A Computer Vision and Robotics Laboratory

Abstract

Using computer vision to support real-time, intelligent robotic manipulation of objects, an australian laboratory is experimenting with a variety of equipment and algorithm combinations. The preliminary results shown in this article are indicative of the type of work supported in the laboratory. As in any laboratory situation, researchers are finding signs of encouragement as well as discovering some of the system's shortcomings as the exploration proceeds. On the positive side, it is becoming increasingly clear that combining range and color image data provides a rich data base for scene segmentation and that this is likely to support experiments in hand/eye manipulation with a fair degree of reliability. Further, it is fairly obvious that actually positioning the manipulator in the scene can permit a degree of feedback to enhance scene analysis and test interpretation hypotheses-something that cannot be provided by remote imaging and range scanning alone. On the negative side, however, it appears that special computing apparatus, including video stream and array image processing facilities, are going to be needed if relatively sophisticated, real-time hand/eye coordination is to become a reality. Knowing this, the laboratory is now aware of the direction it needs to move in. 34 references.