Robots with Kick
Georgia Tech's Tucker Balch says Microsoft Robotics studio could revolutionize his field.
Tucker Balch has seen his share of bad robots. The associate professor in Interactive Computing at Georgia Tech is also general chair of the RoboCup 2007 event, scheduled to be held at the Georgia Tech campus in Atlanta, Ga., in July. The annual tournament attracts teams from across the globe to compete in a series of soccer matches between competing robot squads.
We spoke with Balch about the RoboCup event, the state of robotics and why the field-thanks to recent software developments-may just be poised for a big leap.
How did the RoboCup idea get its start?
Hiroaki Kitano at Sony created a robot soccer competition in Japan [between] 1994 [and] 1996, with the first competition in 1997. That competition became RoboCup. There was also a strong Korean league called FIRA that was created at about the same time.
We will have about 2,300 participants at RoboCup 2007. That amounts to about 380 teams from 22 countries. We are taking over the campus of Georgia Tech. It will be like a mini-Olympics.
Why choose a soccer match?
Soccer offers a lot. Perception of the field, following the ball, balance. If we can do all those things, then all those technologies have
a direct mapping back to practical robots.
If it can kick a soccer ball, hopefully it can do something like go to the sink and [wash] the dishes.
Each year we make the competition a little harder in some way. And we hope that those tweaks to it drive the science.
We used to have walls around the field because the robots weren't so good at controlling the balls. So one year we decided, we're going to get rid of the walls. The teams cried bloody murder, but we stuck with it. They had a problem for a year but sure enough, they got better.
Each year we change all these leagues just a little bit. But the eventual goal is to have the rules be just like human soccer.
So how good are they at this point?
[The small size league] is like high-speed hockey. The ball is the size of a golf ball and the robots are six inches in diameter with wheels. That ball just snaps around.
They can pass to each other and receive the passes, and when they score-the goal is made out of wood-you hear this "whap!" The balls travel at four meters per second and the robots at two meters per second.
The big news, I think, will be about how capable the humanoid robots will be this year. Two years ago it was not very encouraging ... The challenge for the humanoid teams was to kick a ball. The humanoid robots looked kind of like drunk little people.
There was one team [at the 2006 RoboCup in Bremen, Germany] that [performed] really well. They just annihilated the other teams. I think we're going to see five or six teams like that this year and it's going to be a very compelling thing to watch.
||"I think we do need a little bit of order in this area. Robotics up until now has been in anarchy. Even if the evil empire brings the order, it will be good."
|Tucker Balch, Associate Professor,
Interactive Computing, Georgia Tech
How difficult is the software for this stuff?
Software is the roadblock now. We have capable hardware, [but] robot software is now very complex and very sophisticated.
Successful robot software must touch every level in a software system, from drivers at the bottom to [graphical user interfaces] at the top. It is perhaps one of the few application areas that requires integration at all these levels.
So developers have to do the entire stack themselves?
There's been no real structure for robotics development. We have to go all the way down to the hardware. For instance, we have to turn motors on and monitor how fast they're turning. On the complete other end, we have to do the AI [artificial intelligence] and plan.
The way this community has been moving along has been essentially all the universities begin by rolling their own [code] and improving it each year. After a little while, you end up with this crusty piece of code that nobody understands anymore.
How does Microsoft Robotics Studio affect this?
Microsoft Robotics Studio is a big deal. Also, there are some fantastic Korean robot kits that are going to explode on the U.S. market, I predict. Robotics Studio is going to run on lots and lots of different hardware platforms, but they're going to provide a uniform software environment.
I think we do need a little bit of order in this area. Robotics up until now has been in anarchy. Even if the evil empire brings the order, it will be good.
What are the implications for developers?
What Robotics Studio is going to do is provide a framework in which manufacturers can write the drivers that plug into Robotics Studio. I think what it's going to do is allow roboticists to work ... to get drivers and have confidence that they'll plug into the rest of your infrastructure.
It's kind of like video card manufacturers writing drivers for Windows. Now I can write that AI code and know that it will drive that motor. It's like a PC that can run Excel and Word, Photoshop, Dreamweaver-you name it. Right now if a software developer wants to create an app, he has to find somebody to do the rest of the stack for him. And even if it's a great app, it only works on one robot. Now we can target multiple robots with an app.
There are some people who are not thrilled with Microsoft providing the standard in yet another area. Until now, pretty much the standard platform in robotics [has been] Linux.
I'm a Linux hacker and I'm glad to have Microsoft step in, because what it's going to do is force the issue on the Linux side at the same time. I think it's good to have a little bit of competition.
What challenges face robot designers today?
One of the hardest things for a robot to realize is it's stuck and that it needs a person to help it, and initiating that request. There's no quick answer, but when a company figures it out it could get distributed across platforms quickly because of Robotics Studio.
Another big challenge is perception. Robot sensors are nowhere near as rich and reliable as the senses humans have. They're always faced with this challenge of "well, my sensors are telling me this but what does it mean?"
So what's next?
The key challenge for selling people robots is they have to be reliable. I think the iRobot Roomba is probably the first reliable robot that does housework.