Hand-Eye Coordination at the
Johns Hopkins Interaction Lab

Our work in hand-eye coordination is focused on developing simple vision-based systems that perform with high accuracy. We've recently been developing a set of vision-based positioning tools and testing them on our Zebra Zero robot.

Suppose, for example, that you wanted to touch the corners of two floppy disks like this:

What's any easy way to do it? If you can measure the distances between the corners in two cameras, and cause that distance to go to zero, the corners will be touching. This type of operation is extremely accurate even if you don't know the relative locations of the cameras and the robot arm. In fact, its accuracy only depends on how well you can measure where the features are in the image.

We have systems that perform these operations. They rely heavily on a visual feature tracking system we have developed. For this particular operation, the "camera's eye" view looks like this:

This same idea extends to a variety of other operations that require accurate orientation or alignment as well as positioning---for example, placing a screwdriver on a screw:

More elaborate operations can be performed by using various types of geometric constructions in an image. For example, here is a construction for guiding a floppy disk into a disk drive:

Comments to hager@cs.jhu.edu

     Other Links

     Back to Top