Page 60 - Modern Robotics Building Versatile Macines
P. 60

40   Modern Robotics


            upside-down pendulums mounted on moving carts. They developed
            mathematical equations that specified the ranges of motion in which
            balance could be maintained.
              Using related ideas, Ralph Mosher at General Electric developed
            walking devices that would be controlled by a human operator. The
            operator would be placed in a harness such that his or her walking
            motions would be translated into powered walking movement by
            the machine. In 1968, Mosher demonstrated his “walking truck”
            for the military.



            Robot Kangaroos

            Building a dynamic walking robot involves several different systems.
            Besides controlling the specific motion of each leg joint, some sort
            of master controller must determine the sequence in which the legs
            move. In turn, these instructions must take account of feedback that
            informs the robot when it is getting out of balance (and in what
            direction). Finally, there has to be a sort of “strategic planner” that
            determines the robot’s desired destination, examines possible foot-
            holds (places to step down), and picks the best route.
              Raibert and his team at the Leg Laboratory began with a very
            simple proof-of-concept: a robot “pogo stick.” It consisted of a com-
            puter-controlled piston that determined how far the robot could stride
            forward without losing its balance and another control that deter-
            mined how much “spring” or “bounce” could propel the leg from one
            stride to the next. By combining legs, Raibert and his team later built
            two-legged (bipedal) and four-legged (quardapedal) robots.
              They got their idea for the robot from one of nature’s most accom-
            plished hoppers: the kangaroo! They studied the structure of the
            animal’s bones and muscles and analyzed its hopping gait.
              One special thing about Raibert’s approach is that it did not
            require exhaustive calculations or some sort of central control
            system. As long as each leg (or coordinated leg pair) was kept in
            balance, the robot as a whole remained in balance. Nevertheless,
            the filmed trials of the robots include blooper reels featuring the
            worst stumbles and crashes of the early prototypes. But progress
            continued. By 1984, a quadrupedal robot could trot across the
   55   56   57   58   59   60   61   62   63   64   65