Compared to the general robotic community, DARPA seems to get it right. The 2004 and 2005 Grand Challenge, and the subsequent Urban Challenge helped define the capabilities and limitations of driverless cars. Now, with the arival of the new “Cheetah” robot, we can see the capabilities and limitations of running robots. The Cheetah set a new record for speed, running about 18mph for an extended period of time.
If you watch this video, you can see that the robot is, well, less “robotic” than the typical variety. Instead of solving positions in a momentum-free world, the programming has created something that is clearly juggling its heavy weight quite nicely.
Equally, we see the limitations. Because the robot has limited senses, it can’t figure out where to place its feet to avoid falling. So, rather than add said sensors, the current robot has long, curved “sleigh runners” that it lands on. This keeps it stable when doing the little jumps necessary to run, but in all probability make it impossible to run on real ground. Recall that a real Cheetah would be able to bound up a very rough slope at speeds higher than this machine. While the robot’s gallop is impressive, there’s no evidence that it is dynamically adjusting – it looks fairly repetitive.
The worst part are the comments on the video, so lame as to defy analysis. One compared the robot to Skynet’s ancestor, strange since Skynet was clearly a mainframe-class, static AI running a defense network. This machine doesn’t look or act at all like that.
The other problem, which will be with robotics for a long time, is power. There’s a heavy electrical cord held by a human hand, and I shudder to think how fast a battery would wear out trying to power that run. It makes me think that anything short of a nuke will fail to create the robots of science fiction. And, if a robot that jumps is tethered to an electric cable, one might wonder why it should try to run at all.
I supposed even the caged (robotic) bird will sing from behind the bars.