Tuesday, June 14, 2011

On wasting money on robots

Does the US government waste money on robotics? Probably it does.  In the past I've complained about "robot bees", and there does appear to be a lot of robotics research which seems to be poorly coordinated and of questionable value.  However, it's difficult to pick winners when it comes to research (otherwise it wouldn't be research), and it may turn out that robot bees are in fact really useful for carrying out inspections of damaged nuclear reactors or collapsed buildings or mines, in places where other types of vehicle wouldn't be suitable.

Criticising the PR2 for folding towels slowly seems pretty lame though, and this sort of childish characterisation or misrepresentation of technology projects is the main reason why on the couple of occasions where I've been asked to appear in mainstream media I've always declined.

Since as far as I'm aware Willow Garage is a private company the criticism here seems to be against the University of Berkley which carried out the aforementioned towel folding. This argument is easy to refute if you think about how many towels and other items of clothing get folded in the US each year, how much time that takes and what the economic value of that time is.  I suspect that it's a really big number.  If you can produce a general purpose textile folding robot at the right price level then this is potentially a huge industry.  Also the speed of folding on this research robot isn't a big issue.
"Computationally,the throughput was primarily limited by the optical flow calculation: the NVIDIA GTX 295 GPU provided a through-put of 1 optical flow computation per 5 seconds, and the processing of each frame requires either one or two optical flow computations, depending on whether a 2-D grasp point detection is made"
This process only needs to be speeded up by an order of magnitude or so for the system to be more practical, and it seems reasonable to believe that this will happen within less than a decade, especially given the amenability of vision algorithms to parallelisation. Also there are many possible alternative ways to calculate optical flow.  Beyond a certain point it makes no difference how fast the vision computations are, because physics sets the pace.

Also the appearance of the Microsoft Kinect sensor, which happened after the towel folding research was done, may make all the above limitations irrelevant.

As far as government prudence goes probably far more substantial savings could be made in the military and security areas, where there seems to be a lot of profligate spending on projects of questionable value even within their particular scope of interest. Is there really any need for ICBMs to be sitting unused in silos for decades, long after the cold war ended? Was it really necessary for the US to build a second smaller version of the space shuttle? Is the US tax payer getting value for money out of the Afghanistan war? All of these things are basically just parasitic costs.

No comments: