Researchers at the EPFL in Switzerland have built one of the fastest robots on the planet. According to a news release, the multi-axis arm can catch all manner of crazily shaped objects in less than five-hundredths of a second. That sounds pretty impressive, an even looks impressive, but what else can it do?
The robot is a fairly sophisticated four-fingered hand connected to a standard Kuka arm. It is trained using a technique called “programming by demonstration,” which basically means imitation by trial and error. The trainer creates sample trajectories for the arm by manually guiding it, similar in a sense to what babies must be doing when they learn to control their own arms. Using an array of cameras, the robot creates a model of the object’s motion. This model includes the object’s flight path, rotation, and whether it is a hammer, racket, bottle, or something else.
One of the potential applications offered for this arm is for collecting space debris. While there would seem to be a bit of a gap between catching 10 mph sporting goods and 20,000 mph space junk, it all makes slightly more sense (from a funding point of view) when noting that there is also a Swiss Space center nearby at the EPFL. More practically, a robot arm with this kind of speed would give even the new Deka arm a run for its money.
Now that robotics researchers are beginning to realize the futility of bottom-up programming in which a near infinity of possible scenarios must be prefigured in software, top-down approaches are all the rage. If the larger world consists of more than a limited set of objects, the real top level must lie significantly beyond mere object identity. But what is the real top level? Perhaps what we humans do is use something more like good and evil to represent various object incidentals.
Perhaps the best way to answer the question of what to do with an incoming ball or hammer is to apply a bit of will and feeling to the problem. Lacking pain receptors, a robot has little skin in the game when it comes to choosing among competing possible trajectories to use. A human, on the other hand, seeks to minimize the penalty of the sudden impact while at the same time maximizing the robustness of the response. In other words, the human wants to bring its large muscles into play for initial speed and rigidity, but also its smaller effectors to precisely decelerate the object in the most pain-free way possible.
Having constraints like these greatly simplifies the problem of how to intercept an object. The question then becomes not how to orient an arm or hand, but getting it to respond in the first place. In the not so uncommon event of stroke or even intentional surgical procedure for epilepsy, patients sometimes walk away with a bizarre affliction known asalien hand syndrome. Not only does the victim of this condition lose control over the hand, but the hand will sometimes do the opposite of what is desired; this frequently becomes the worst thing imaginable in any given circumstance. Alien hand syndrome falls under the province of neurology, which in practice, is chiefly concerned with converting the strange into the obvious. While for many, the height of personal absurdity would have to be our left hand randomly choking or undressing us while the right hand tries to restrain it. For a robot, we might expect this to be a familiar bug we dial out in much the same way we now tweak a servo loop.
Clearly, there is much to be added to the robot model before it is capable of reducing a 10-degree-of-freedom joint configuration space into a cascading tree of binary good versus bad decision points. Drawing such a distinction may now seem as strange as assigning gender to names for objects in many of the world’s languages. But even for things like space debris, the most important thing about it that we need to know may be the split-second decision of whether to avoid, destroy, or keep it.
