home · mobile · calendar · colloquia · 1996-1997 · 

Colloquium - Smith

Vision-Guided Robotics
Department of Computer Science and Engineering, University of Colorado at Denver

Recent work combining robotics with vision has emphasized an active vision paradigm where the system changes the pose of the camera to improve environmental knowledge or to establish and preserve a desired relationship between the robot and objects in the environment. Much of this work has concentrated upon the active observation of objects by the robotic agent, leaving interaction as an issue for future work. Since robots are designed to accomplish tasks, many of which involve manipulation of objects, our approach emphasizes techniques for the real-time control of robots via computer vision. Rather than simply observing, we design systems that interact with their environment in a reliable, accurate and robust way. Further, we attempt to use visual feedback control throughout the specified task, leading to a paradigm we have named vision-guided robotics,

We present an overview of the area of vision-guided robotics and discuss the issues involved with using computer vision as a sensor for real-time robotic systems. The resolution of these issues and experimental results from developed systems demonstrate the effectiveness of computer vision combined with appropriate control laws when applied to real hardware. We also discuss adaptations of the basic system software, demonstrating the applicability of our techniques to other areas of research, most notably Intelligent Transportation Systems.

The culmination of this effort is the development of an automated grasping system that uses computer vision throughout the alignment of the gripper, the reach toward the target, and the closure of the gripper to grasp the target. Contrary to previous efforts in vision-based grasping, our system does not require extensive calibration, stereo cameras, nor any blind reach component (i.e. operation without visual feedback). Furthermore, the object is not required to remain stationary during the grasping motion of the manipulator.

We also present extensive experimental results demonstrating the effectiveness of the grasping system on multiple grasping trials using both static and moving targets with random orientations, locations, and depths.

Refreshments will be served immediately before the talk at 3:30pm.
Hosted by Dirk Grunwald.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)