(continued from previous page)
So what technology do you think needs to be perfected before a humanoid robot like C3PO becomes a reality?
Mason: The easiest way is to put a human in a machine.
You mean use the human machine as a model?
Mason: No, I meant literally. I was joking. Oh, you mean without the human?
It's interesting. What was the name of the planet where R2D2 goes?
The Ewok planet?
Mason: No. Where he goes with Luke to the jungle, the swamp to see Yoda?
Tatooine? No...(Correct answer is actually Dagobah.)
Mason: Well, anyway, you see R2D2 following Luke and Yoda is leading them back to his tree stump. You're looking at R2D2 and wondering how he's navigating this forest floor with vines and everything with those little wheels. I don't think there is any technology that can handle that well yet. C3PO is more plausible because he's based on a human chassis.
The one thing we learn over and over in robotics is everything is much harder than anyone thought. We thought chess would be a great challenge for AI and now we have chess machines that can beat every human, or almost every human. Turns out, building a machine that can manipulate the chess pieces as effective as a human is way harder. I mean, sure, we can build them for a specific set of pieces and a specific chess board, but not where you can show up with any chess board and it can take the pieces and sort them out and start playing.
Ultimately we've discovered that mopping a floor is harder than playing chess. I guess that means janitors are as intelligent as anyone else. The distinction between them and anyone else is negligible.
What do you see as a less popular or unexplored area that you would like to see more research done in?
Mason: Theoretical underpinnings. It's easy to motivate research when it's close to application and when it's close to a machine. Machines are inspirational. You see a machine and right away, not knowing what it can do, you project capabilities they don't even have. It's much more difficult to attract funding for longer range research. Ultimately the impact might be as great because it can apply more broadly across different types of robots.
What's the biggest AI achievement so far?
Mason: One of the really exciting things going on right now is the development of statistical methods, machine learning techniques especially in robotics. But that's one among many. I'm trying to think of what else.
Well, what's the most interesting work being done at CMU?
Mason: Hmm. That's really hard to say. There are a lot of interesting things. One of them is an image understanding system, being offered on the Web as Fotowoosh. You give it a picture and it gives you back a virtual reality three-dimensional structure that you can fly through. It can figure out the image...I've seen it work for street scenes and seascapes and outdoor things.
CMU is in this year's DARPA challenge and it's going to be in an urban setting instead of the dessert like before. It seems like that would be more difficult.
Mason: With the DARPA challenge, the first year was very challenging. But it's hard to do a comparison because we are now taking back machines that are radically better. But the urban grand challenge is definitely more interesting and challenging in that you are dealing with other vehicles.
You know. I think another one of our most interesting things is with origami.
OK. So, tell me about the origami robot. What sorts of practical applications do you see that type of technology eventually having?
Mason: What do you mean? We did it 'cause we really love origami. Come on, don't you envision them in every mall in America making 1,000 cranes a minute?
I'm only kidding. We do it 'cause we're interested in the principles and so far the most automatic program research in robotics manipulation does harder things. Literally you have it easier to manipulate hard things than soft things like paper. And so, for us, extended out planning algorithms for paper was a challenge. We think origami will be for manipulation what robot soccer is for mobile robots: a great challenge task that can inspire and challenge researchers for the next 50 years.
1 commentJoin the conversation! Add your comment