Vision and Robotics Research

at the University of Rochester

Projects Mobile Algorithms VR

Research in robotics and vision at UR spans a wide range of topics and applications, however one unifying aspect of the various projects is the focus on techniques requiring no or little calibration and thus are suitable to use in natural everyday environments. For example Randal Nelson and Raj Rao are each working on different variations of appearance based recognition which does not need any prior CAD models of the objects. Olac Fuentes has devised and implemented a way of grasping and manipulation a wide variety of objects with a Utah/MIT hand without the use of any object models. Martin Jägersand has developed a method which can learn Visual-Motor manipulation models on-line during the manipulations, instead of requiring a-priori calibration, and shown how to use the method to solve several everyday manipulation problems. Chris Brown and several students are working on vision for a mobile robot. Another benefit is that uncalibrated manipulation techniques lend themselves to better man-machine interfaces, where the robot is instructed for instance by the user pointing at objects (while wearing a "cyberglove" or with the mouse in a picture), or by drawing a "virtual sketch" in a "Mac draw" like program of the desired manipulation. This focus on uncalibrated methods is in contrast to most robot vision systems currently in use, where a carefully engineered environment, and an accurate a-priori model are required, and the manipulation task is described in how many millimeter the robot needs to move in a fixed world coordinate frame.

Another main focus of our work is to use advanced vision and virtual reality techniques in psychophysics experiments to learn about the human brain and the human visual system. In one project Mary Hayhoe, Dana Ballard and students study human visual attention and short term memory in a virtual world where for instance objects can be made to appear and disappear, or change color in real time controlled by the gaze direction of the human subject. Using these methods they have found that the human uses surprisingly little internal memory during typical manipulation tasks, and instead accesses information in the world as it is needed. Kyros Kutulakos and Jim Vallino use computer vision techniques, requiring only minimal calibration, to augument real-time video streams with graphics simulated or real objects.

Research Projects:

Calibration free robotics using intelligent sensing.

Augmenting and synthesizing images and movies.

Recognizing objects,faces, and activities without geometric models.


Visual navigation and mobile robotics.


Computer vision algorithms.


Virtual reality applications.

Attention, psychophysics and models of human vision.

Some related courses

Other links:


We were selected

"Cool Robot Of The Week"

by NASA on Jun 8 1998

Martin Jägersand ,