Robots don't always see eye to eye. Sometimes they don't even use the same techniques for viewing.
Companies are taking different approaches to computer vision systems, one of trickiest problems in robotics, according to a whirlwind round of interviews at the RoboDevelopment Conference taking place in San Jose, Calif. But it's only one of several forks in the road for robot developers. The "humanoid or not" debate continues to rage, and companies still question whether the strongest market for robots will be with consumers or industrial customers.
First, the visual debate. In computer vision systems, cameras feed data about the surrounding environment into the robot's processor, the processor interprets the image and acts. It's relatively easy when you have stark contrasts--find the circle in a field of squares--but the real world is filled with ambiguities.
RoboRealm is trying to fine-tune two visual techniques. It has developed software that lets robots recognize shapes--squares, triangles, circles, and so on--as well as color.
By contrast, Canada's Braintech is looking to exploit visual textures. Instead of looking for a color, it will look for a picture that matches something in its databases. In a demonstration at the conference, the company rigged up a Lego robot to find an orange juice container and then put it in a recycling bin. Rather than search on the orange color or a large rectangle (the shape of the box), it looked for the label, which showed a pile of cut-open oranges.
Once it grabbed the juice box, it headed toward the familiar recycling symbol. The goal can even be obscured--there was a Planter's can of mixed nuts in front of the recycling sign. As long as about half of it shows, the robot can identify it, said Robert Sim, senior robotic vision scientist at Braintech.
The human or machine debate, meanwhile, was thought to be settled a few years ago when Sony killed off its Aibo and Honda owned up to slow sales for the Asimo. People were buying iRobots at the same time. The general conclusion was that the future lay in producing functional robots, rather than entertainment devices, and machines with personality be dammed.
Entertainment robots, however, suffered from two problems. The robots didn't come with a back story to make them interesting, said David Hanson, founder of Hanson Robotics. To that end, Hanson will come out in 2009 with Zeno, a robot boy that responds to questions and interacts with people. He's supposed to be from a robot camp in the future and kids will teach him about life on Earth.
The responses are less scripted than in previous humanoid robots, which should make it more engaging. Also, Zeno will cost only $300. Hanson is working with Massive Software, the guys out of New Zealand who did the artificial intelligence for the imaginary characters for the Lord of the Rings movies.
Another humanoid robot backer is Anybots, which makes life-size human-like robots designed to perform household tasks. Because factories and homes are built to human scale, it makes sense to have robots with arms and legs in the same places as humans.
On the other hand, you have Willow Robotics, the new robot start-up making household robots. Company representatives said they don't care about human characteristics. The thing is made to pick stuff up. Most companies, such as Israel's Galileo, are in the non-humanoid camp.
Finally, there is the legs versus wheels divide. Legs are complex and bound to fail, said many. Wheels, or tank treads, can get a robot to go almost anywhere. But Karl Muecke, a graduate student at Virginia Tech who built the Darwin legged robot, points out that legs are good for stairs and other terrain. Bipedal locomotion remains a key area of research at the school.