WORKSHOP on VISION FOR ROBOTS Sunday August 6th, 1995 The Westin William Penn Hotel, Pittsburgh, PA 15219 in conjunction with International Conference on Intelligent Robots and Systems (IROS-95), Aug 7-9, 1995. General Chairman: Katsushi Ikeuchi (Carnegie Mellon Univ.) Program Chairman: Avi Kak (Purdue Univ.) Program Committee: Minoru Asada (Osaka Univ.) Peter Allen (Columbia Univ.) Ruzena Bajcsy (Univ. of Pennsylvania) Akio Kosaka (Purdue Univ.) Martial Hebert (Carnegie Mellon Univ.) Seth Hutchinson (Univ. of Illinois, Urbana-Champaign) Charles Thorpe (Carnegie Mellon Univ.) It could be said that the proof of the pudding in building a computer vision system is to demonstrate it on a robot. After all, in order to make a robot do anything useful through the use of its vision sensors, the vision system must work with some degree of competency and robustness, not to mention the fact that in addition to scene interpretation the vision system must also yield information for the pose calculations needed for subsequent robotic manipulation. Over the last decade, a number of research groups have actually demonstrated robotic vision systems for both the arm robots and the mobile robots. For the case of arm robots, researchers have demonstrated bin-picking of non-polyhedral objects using 3-D vision systems. And, for the case of mobile robots, researchers have demonstrated navigation modules using monocular and binocular vision. The aim of this workshop is to bring together people who have some experience with the integration of vision systems with robots, both the arm robots and the mobile robots. Discussions at the workshop will focus on defining more sharply the current state-of-the-art in the design of such systems. We will also try to delineate the next frontier of experiments for this kind of research. More specifically, in addition to reviewing the progress in the design of integrated robotic vision systems, the workshop will address a host of technical questions that appear highly relevant to the field. For example, while for bin-picking applications we have had great success with 3-D vision, nothing comparable can be said for 2-D vision. Can any lessons learned from 3-D vision be applied to crack the problem of 2-D vision? For the case of mobile robot in indoor environments, is it better to use precompiled models of the environment, or should the robot construct such models using its sensors? What are the geometry vs. topology tradeoffs for the representation of models for mobile robot navigation in general? What are the best algorithms for pose calculation? Where do we stand in bridging the gap between the closed-loop systems for visual servoing and model-based systems for scene recognition? What about the use of motion cues? How successful have been the concepts of purposive vision? PAPER SUBMISSIONS: Four copies of the full paper including figures and drawings (double-spaced, not exceeding thirty pages) must be received by Mar 1, 1995 to the Program Chairman. IMPORTANT DEADLINES: Submission of papers: Mar 1, 1995 Acceptance notification: May 1, 1995 Submission of final camera-ready papers: Jun 1, 1995