A good robot is hard to find -- or build

As competitive research teams stumble at NASA challenge, a look at the struggle behind robotics

WORCESTER, Mass. -- In the first two days of the NASA robotics challenge this week, every team failed.

Every. Single. Team.

NASA officials and roboticists were disappointed, but not shocked, at the less-than-spectacular results at yesterday's NASA robotics challenge -- even though researchers from the likes of Rensselaer Polytechnic Institute, Worcester Polytechnic Institute (WPI) and Oregon State University had worked on the fully autonomous machines.

It's not that robotics is particularly new territory. People by now are accustomed to iRobot's Roomba robotic vacuum cleaning our floors. Robots help assemble automobiles and smartphones. A humanoid robot is even working on the International Space Station, while robotic rovers are exploring Mars, looking for signs that the planet has ever been able to support life.

Those robots, though, aren't fully autonomous. There are humans calling the shots behind the scenes.

The robots that 18 teams from the U.S., Canada, Estonia and Mexico were looking to show off at a course on the WPI campus here this week are simply a different breed of machine.

In its third year, NASA's Sample Return Robot Challenge is focused on advancing fully autonomous robots. Researchers have built the hardware and software for robots that can traverse an area the size of one and a half football fields, find objects and retrieve them.

At least, that's what they're designed to do.

NASA is looking for technologies that can help its engineers build more advanced robots that will eventually be used in deep space, such as on Mars or on asteroids.

These machines have been designed to start up, figure out where they are and then proceed to do their work -- all without human guidance. Sounds simple, but it's not.

"There are so many ways to fail at this," said Jascha Little, a mechanical engineer on Team Survey, a Los Angeles-based group of individual researchers that competed unsuccessfully on Thursday. "We're still experimenting. This is nothing you can buy. You throw together parts and software libraries and try to make a system out of it.... It's just that hard."

After three years of work on their robot, after writing about 10,000 lines of code and sometimes spending as much time on this "spare-time project" as they do on their full-time jobs, Team Survey's robot didn't even make it off the starting platform.

They weren't alone in their frustrations.

Of all the participants, which included teams ranging from groups of individual researchers to university-led teams, a handful never made it off the starting platform.

But two teams nearly had a successful run.

One robot built by a group from Estonia found a sample on Wednesday, the first day of the challenge. But instead of retrieving the object, the robot accidentally ran over it.

Another team from West Virginia actually found an object and was able to pick it up. However, the robot didn't recognize that it had captured the objected and dropped it. It tried again and again, dropping it each time. By the time the robot stopped trying, the object had been driven into the ground.

"It's a problem of sensor fusion," said Craig Putnam, an adjunct computer science professor and robotics instructor at WPI. "You have lots of different kinds of things you're sensing and you have to fuse that all together to get the big picture. There's vision, odometry, the tilt of the ground, colors and shapes.... It's just a really hard problem."

At NASA's Sample Return Robot Challenge, held this week in Worcester, Mass., robotics experts discuss why teams have had a difficult time succeeding with their autonomous robots.

1 2 Page 1
Page 1 of 2
Bing’s AI chatbot came to work for me. I had to fire it.
Shop Tech Products at Amazon