MINSKY@OZ.AI.MIT.EDU (01/23/88)
We did make a ball-catching robot in the late 1950's. Richard Greenblatt and William Gosper were involved with it. We considered ping-pong but concluded that our mechanical arm - then an AMF versatran machine - would be too slow. Somehow, though , a rumor spread that we had a project to play ping-pong. The only substance was that Gosper did make an attitude-controllable paddle that could be attached to the arm in case we were able to speed it up. It should be mentioned that Gosper was - and presumably still is - a master level ping-pong player. The rumor about the ball-catcher trying to catch me was true, however. It would try to catch anything it could. We built a railing around it. It was no good at catching people because the algorithm was: find anything that moves and extrapolate the appropriate parabola for free flight. But it sure was dangerous being around it. Our pincer-like mechanical hand was also to slow to catch the ball, given the low accuracy we were getting from our TV camera. Greenblatt finally attached a rustic-looking straw cornucopia baskets to the arm - you know, the sort of thing shaped like a curved horn with the flared end up. Most visitors assumed that we were simply trying to be quaintly anachronistic. The sad fact is the the cornucopia was about the only thing that worked. We had tried all sorts of little cups and pails, etc., but the ball would usually bounce out of them. The ball was too dumb, though, to figure out how to get out of the cornucopia. Later, as we got into the problems of robotics - for example, the problems Fahlman faced with the BUILD program - we decided to leave the problems of high-speed but low-level robotics aside. For the same reasons we stuck with stationary rather than mobile robots.