Some background:
The academic robotics community values new ideas above applications of existing ideas. I agree with this, to some extent. It’s hard to solve a novel problem, and it’s obviously critical to the research baseline of Are Things Getting Better. Applications are critical too, but somewhat easier – in theory, it’s just coding up an idea on a physical system.
“In theory, there is no difference between theory and practice. But, in practice, there is.”*
Pretty much as soon as you start trying to apply your (or someone’s) neat new ideas to actual physical robots, you discover that your simplifying assumptions are too simplified, or your algorithm doesn’t cover that one case that comes up 90% of the time, or everything is theoretically perfect but it just doesn’t work because your perfect motion planning makes the robot overheat.
Building and programming a robot that can just sit down and play chess with a human (vids: robot, me talking), with whatever pieces you care to use and without special boards and whatnot, took months. In the actual chess competition, we destroyed the other entrants. And the question I got – at the competition, and on the resulting paper – was, “I don’t understand. What’s novel?”
 |
| I feel bad about how wordy this is. So, lookit! A robot playing chess with an adorable child! |
On the one hand, a robot that actually plays chess with actual humans, duh. On the other hand – the getting-hired-in-academia hand – nothing. We made existing stuff actually work on a new platform, with all the engineering headaches and discoveries of simplifying limitations and so on that that implies. No really new ideas. All we did, with months of effort by a lot of really smart people, was take a lot of existing ideas and demonstrate that they actually worked, or in some cases, didn’t, and used them to build a system that did something that hasn’t been done before.
That’s nothing.
Me, I’m too junior to have an opinion on how stupid that is. That’s an opinion I can have after getting tenure, maybe. But enough tenured people got fed up enough with it, or wanted to see Honolulu enough, or whatever, to form ISER, which is basically the “Look, my robot can do things!” conference.
So the conference was awesome. There were papers about using robots to control invasive fish populations (by tracking the fish, sadly, not by zapping them with underwater lasers, sadly); about using pictures of scenes to figure out what parts of rooms look navigable (the parts with people in them); no using a robot to make cookies; on following directions... all kinds of brilliant stuff. Some of it was vision-only, or quite theoretical. The only thing all the talks had in common was that someone actually tried something to see if it worked.
(Well, that and being presented by people whose universities could afford to spend insane amounts of money on hotels, food, and conference registration fees for a second-tier conference. No scholarships at ISER. Still, it was a lot of fun.)
* This quotation, which exists in several forms, has apparently been attributed to Yogi Bera, Jan L. A. van de Snepscheut, Albert Einstein, etc...
4 comments:
Isn't taking existing ideas, rather than creating new ones, and forging them into something that actually works in large part the purview of "industry?" I am uninformed on this point so I ask the question as the most direct route to become informed. In cancer research, academia highlights the molecular targets and industry shoots at them (makes drugs). What you're lamenting is the lack of a PhD level trade school equivalent to Med School.
So this is where you're blogging now. Thanks for the nifty post about a subject I know nothing about.
Hi S! Okay, first note that ALL of this is surrounded by tags.
That said: That's actually exactly the response I would probably get from almost anyone in academia. And to some extent, yes, that IS the perfect-world divide. In practice, though, it's more of a spectrum? There are industry research labs, and there are experimental-robotics academics. Both to do, mm... applied stuff that's still quite a ways from being industrialized? So there's substantial overlap. For example, one of the research groups at UW CSE spawned Travelocity (now Bing Travel), Zillow, and a few other very successful startups, the same way Captchas and Duolingo came out of academia. On the other hand, Intel, Google, and MSR publish heavily on their further-out stuff.
The reason it annoys me is because I feel like people on both sides of the fence pretend it's a much clearer split. No company was going to pay for that group to make a chess-playing robot to attend a AAAI competition (even though Intel paid for the robot), and how well we did really defined the future of the competition. In academia, I've gotten a lot of attention from those videos. So when people act all "but this is (eww) applied robotics," I feel like they're perpetuating an ivory-tower myth.
Iunno. Anyone else want to weigh in?
Hi Cathy! :-)
Read the response to SG, then - it's a more complex subject than I'm making it out to be in the original post.
Post a Comment