Page 155 - Modern Robotics Building Versatile Macines
P. 155

RADICAL ROBOTICIST   135


            Robotic Vehicles


            Moravec moved to Carnegie Mellon University (CMU) in Pittsburgh
            in 1980 as a research scientist. He was promoted to senior research
            scientist in 1985 and principal research scientist in 1993. For his
            first project at CMU, Moravec began to develop a successor to the
            Stanford Cart: the CMU Rover. Unlike the Stanford Cart, the Rover
            carried a dozen or so computer processors on board, although the
            heavy-duty image processing was still handled by a minicomputer
            on a remote link. In addition to a TV camera, the Rover included
            infrared and sonar sensors. The robot was about three feet (1 m)
            tall and weighed 198 pounds (90 kg). New image-processing arrays
            made picture analysis about a hundred times faster than with the
            old Cart.
              In 1984, Moravec and his team began a contract to develop
            a sonar navigation system for Denning Mobile Robotics. Since
            the sonar provided distance information but could not localize
            an object within its 30-degree-wide beam, Moravec and gradu-
            ate student Alberto Elfes devised a different approach to avoiding
            obstacles. A three-dimensional grid around the robot was used to
            plot the possibility that an object may exist in a given cell. The
            result of successive “pings” generated a sort of probability map,
            which was later extended to combine sonar and visual data. Finally,
            an algorithm was developed by which the robot could improve its
            picture of the world by comparing it to a simulated map. Essentially,
            the result of all this processing was that the robot did not try to
            avoid obstacles, rather to determine a route that was safely free of
            them. By the early 1990s, the availability of new supercomputers
            such as the CM-5 “Connection Machine” improved the accuracy of
            this grid-based navigation system steadily, and progress continued
            through the decade.
              Moravec’s Carnegie Mellon lab also worked on a series of self-
            driving vehicles. The earliest version, the 1984 Terragator, could roll
            along jogging trails at about three feet (1 m) per second. Sometimes,
            however, the remote-control computer would confuse a tree trunk
            with the road and the Terragator would try to climb it!
              Navlab, the first of a new series, begun in 1986, was a big blue
            truck full of bulky computer gear. It used algorithms to try to pick
   150   151   152   153   154   155   156   157   158   159   160