How contests lead to robots that jump
Nice article over on the Robotics Trends
website about the upcoming DARPA Grand Challenge
. A key point the article makes is that DARPA had little success in getting standard military contractors to design a useful autonomous robot, and decided to have the contest to bring new blood into the field. They appear to have succeeded. A year ago, nobody thought it was possible for an autonomous robo-car to make the trip from Los Angeles to Vegas at all; today some teams are predicting that their systems can navigate at 40 mph.
In addition, the article quoted individuals entering the race who had been involved with robotics for decades, who said that the race had really got things going again. This attitude really fits with the “long winter” model for robots discussed by Hans Moravec. In Moravec’s recent book, Robot, he notes that artificial intelligence researchers used computers with the same power from the late 1960s through the 1990s. During the earlier period, high initial hopes gave them access to the fastest computers available. During the later period, disappointing results caused downsizing of computer expenditures. The end result was constant computing power at the level of a 1993 Macintosh – interesting, but probably below the threshold for useful robotics.
It is becoming clear that the “long winter” applies elsewhere as well. Hobby robotics had a promising initial start in the early 1980s as the first microcomputers appeared. But while PCs took over the world, hobby robots went nowhere commercially – and the computing power used by the typical hobby robot in the late 1990s was equivalent to that used in the early 1980s. Today, with the appearance of low-power mainboards with 1GHz processing, hobby robotics is just starting to see its first spring.
One sure sign of a robotic spring is the emergence of midlevel projects – larger than hobby robots but smaller than big commercial efforts like the Asimo. This mid-range (spending anywhere from a few hundred thousand to a few million on their projects) is very well represented in the DARPA Grand Challenge. This is one of the key values of contests of this sort – it can motivate and partly fund mid-level projects which will be absolutely necessary to bring on the robotics revolution. Consider the parallels with aircraft. The “hobby” level 100 years ago consisted of making small wooden models powered by rubber bands. The institutional level was represented by big (ultimately failed) projects at universities. The mid-level was a small team like the Wright Brothers, who ultimately accomplished real powered flight. During the early era of airplanes, numerous contests spurred the “mid-level” (Lindburgh) and advanced the field faster than hobby or institutional level research.
Contests are the surest way to get to real robots that jump. (I don’t care if there are wheels or legs as long as they can jump). Already, we see bragging rights in the DARPA challenge going to the most sensors rather than the most clock cycles – a welcome trend, since the Moore’s Law equivalent for robots is sensor density on a dexterous body.
One final note – DARPA has announced that there will be a spectator section – check the spectator page for details. This is the last, but not least, reason for contest – getting the public to see robots outside of Hollywood fantasy and inside a factory welding car bodies. Here’s hoping ESPN starts carrying the Grand Challenge or similar contests as a real sport!