Page 107 - Modern Robotics Building Versatile Macines
P. 107

THOUGHTFUL ROBOTS   87


            of AI in the Stanford Artificial Intelligence Lab (SAIL). He also
            joined in the innovative projects being conducted by researchers,
            such as Hans Moravec, who were revamping the rolling robot called
            the Stanford Cart and teaching it to navigate around obstacles.
              In 1984, Brooks moved to the Massachusetts Institute of
            Technology. Like Stanford, MIT was a burgeoning center of AI
            research and robotics. There he would undertake more than two
            decades of innovative research. Much of it would stem from a
            decision to begin thinking about artificial intelligence in a dif-
            ferent way.



            The Challenge of Vision

            For his Ph.D. research project, Brooks decided to tackle one of the
            toughest challenges in AI: creating systems that can identify and
            “understand” objects in a three-dimensional environment. First,
            Brooks and his fellow graduate students needed a robot chassis on
            which to mount the cameras and other gear. Fortunately, a high
            school student named Grinnell More and two of his friends had
            started building a simple steerable robot called VECTROBOT.
            Brooks not only bought one of the machines but also enlisted Moore
            as an informal research assistant.
              They equipped the robot with a ring of sonars (adopted from a
            camera rangefinder) plus two cameras. The cylindrical robot was
            about the size of R2D2. Since this was still the 1980s, there were
            no small computers powerful enough to run the AI software, so the
            robot was connected by a cable to what was then a powerful mini-
            computer.
              One reason why computer vision is so difficult is because the
            appearance of an object can change radically depending on the angle
            from which it is being viewed. A human has no problem knowing,
            for example, that an upright glass and a glass lying on its end is the
            same object. Computers, however, have to use complex mathematics
            to identify objects and their relative positions. These calculations
            are so intensive that a robot such as Stanford’s Shakey, from the
            1960s, had to “think” sometimes for hours before being able to
            move around an obstacle in the room. It took that long to create or
   102   103   104   105   106   107   108   109   110   111   112