The Future of Robotics, Part 1: Why haven't we built C-3PO yet?

Science fiction has been teasing us with robots for decades. From Isaac Asimov's tome-like novels filled with law-bound robots to the helpful or menacing 'bots on the big screen, many of us grew up with the idea that a robot-filled future was an inevitability rather than a likely outcome.

So where the hell are they? True, in a way, we do live in a robot-filled future. Robots are teaching classes in Japan, and you could wake up as a cyborg yourself in as little as three years.

Still, we want a real walking, talking, thinking robot. A heavy-duty metal-man such as C-3PO would do nicely. He's articulate, expressive and can operate in ways that approach a human's aptitude. Will we see anything like the gold-colored protocol droid in our lifetimes? In the first of a three-part series, we talk to James Kuffner, an assistant professor at robot-friendly Carnegie Mellon, and get deep into the heart of the matter with what it would take to construct a robot like C-3PO.


Robot-Roundtable-James-Kuffner.jpgJames Kuffner, associate professor at the CMU Robotics Institute, is a specialist in the field of motion planning. His work at Carnegie Mellon University is all about getting robots to move around more efficiently — something that C-3PO did with far more ease than today's robots.

The core of his research is his team's approach of building a "search-based" AI, or giving a robot a brain like Google, that can rapidly search through its knowledge for the appropriate entry.

DVICE: What do you do that would help us build a protocol droid today?
Kuffner: Basically, my group is focusing on motion generation — in particularly motion planning. We're trying to apply these search-based AI techniques to practical problems: loading a dishwater, walking up stairs and so on. Imagine the brain of a robot being driven by a search algorithm that allows it to learn things very quickly — like how it could bend its leg if it wanted to step over a branch, and then be able to reason about its connectivity to the space around it and employ the right actions.

It turns out this search-based AI is pretty general. You can give it general parameters like how long its legs are and how far it can bend and then it could quickly reason how to use its legs.

James-Kuffner-humanoid-robots.jpgA gallery of some of the humanoid robots James Kuffner's research affects

How does a search-based AI work?
Just like with the Internet and Wikipedia and databases like that our search-based approach allows us to keep building. At first engineers had to hard code actions and responses into robots and build a database each time — and who wants to do that? The goal is for a robot to be able to search back through its memory and know what it's learned and what the robots before have learned and continue to pass on that searchable database onto the next robot, so that it keeps learning.

Every human child has to learn to pick up a cup and do it over and over to learn how much force is required or how slippery glass is. The great thing about a robot is that it learns what actions to take and what not to do — and if I can just copy that and propagate it and disseminate to future robots, then future robots will already know what to do with a cup.

Kuffner-thinking-robot.jpgWhat benefits are there to giving a robot a humanoid shape?
Well, if the only thing I wanted my robot to do is mow my grass, then I could just stick a radio-controlled receiver on it and drive it around. But the whole idea is that if we design a robot that has a human form, then it can use tools and navigate stairs and buildings and do things that we have designed for the human form.

I remember we were demonstrating our robot to a bunch of Japanese school children and they all came in and bowed to our robot - which isn't something you'd normally do and we weren't ready for that. But the important thing is that this humanoid form allowed our robot to interact in a specific emotional way.

C-3PO was just as much of a humanized character as Han or Luke. Is a 'bot like him an improbable dream?
The idea is he really looks like an English butler-robot who is able to give advice and translate and be a really well-mannered robot. In some ways a lot of robot researchers are threatened by Hollywood because Hollywood shows all of these very advanced robots and it really raises expectations on what a robot should be able to do. You often hear, "Gosh, well your robot really isn't all that interesting!"

I believe the next big thing is robots — just like with computers in the '50s that were bulky and unreliable and only the biggest research institutes could have one, robots are like that now. I'm betting within maybe, you know, the 2020s or so we'll be able to mass produce for a market demanding robots and I believe people will see robots transform more in a way that they've come to expect. At least I hope so. I could eat my words of course, but I'd really like to see robots that early.

James-Kuffner-favorite-robots.jpgWhich robot is your fave?
Actually, because I grew up watching C-3PO and R2-D2, that pair is my favorite. We've actually got a CMU robots hall of fame and they were inducted two years ago. But that peaceful design is something I aspire to. I think in the US we have a fearful relationship with robots because things like the Terminator movies and the idea that technology will turn on us.

I spent quite a bit of time in Japan and there it's much more natural to accept robots. It's an entirely natural thing to think that robots will be our friends and that they can help us.