Mathematicians Attempt to Design More Human-Like Robots

Institute mathematicians have made major strides in the continuing quest to transfer elements of human intelligence -- particularly vision and control of arm-movements -- to robots.

One of the main difficulties in endowing robots with "vision" is teaching them to recognize three-dimensional objects, which can be observed from an infinite number of viewing angles. In an approach developed by Prof. Shimon Ullman and Dr. Ronen Basri, a 3-D object is represented by a handful of 2-D snapshots. A set of mathematical calculations is then used to combine the snapshots that best coincide with novel views from any desired angle.

This method has recently been successfully tested on computer models by Basri and Dr. Ehud Rivlin, who plan to try it on actual robots by storing in their memory 2-D pictures of a room containing contours of various objects, such as furniture and wall paintings. The robot, equipped with a camera, would then determine its location in the room, or reach a designated target, by taking pictures of various surrounding objects and then matching them against the contour bank in its memory.

In related work, Dr. Shimon Edelman collaborating with Prof. Heinrich Bulthoff of the Max Planck Institute for Biological Cybernetics, has shown that the human brain itself may recognize 3-D objects in this very fashion ? by comparing its 2-D image to a series of memorized "snapshots" of previously seen objects. The scientists have provided evidence that recognition takes place when the rain, through interpolation, selects the "snapshots" that most closely approximate the new object.

Prof. Tamar Flash focuses not on vision but on the brain's control of arm movements. She combines laboratory measurements with various modeling techniques to characterize the arm motions of healthy people. Furthermore ? together with Dr. Rivka Inzelberg of the Tel Aviv Medical Center and Prof. Amos Korczyn of this Center and of Tel Aviv University ? Flash studies the movements of people suffering from Parkinson's disease and other neurological disorders in order to pinpoint how they differ from the norm. Her research is also aimed at developing advanced motion control and planning schemes for robotic arms.