Robot gets a driving lesson for DARPA challenge

WPI robotics researchers are working toward competition finals and a $2M prize

WORCESTER, Mass. -- A vehicle drives toward a disaster site, robot at the wheel. The robot stops the car and then steps out to walk toward the disaster.

That's not scene from the latest sci-fi movie, it's what scientists and military leaders hope to see next year when robotics teams from around the world compete in the DARPA robotics challenge finals.

With the last challenge just eight months away, the various finalists - including teams from Worcester Polytechnic Institute, MIT, Virginia Tech and NASA's Jet Propulsion Laboratory -- have been working to get their robots ready to take on tasks ranging from opening doors to using a drill, climbing a ladder and turning valves.

wpi finals

These are tasks that the robots had to tackle during their last challenge. While this time around the robots will need to act more autonomously, most of the tasks they face aren't new.

DARPA did throw a bit of a wrench into the process, though, adding extra difficulty to a trial that already is pushing the boundaries of autonomous and humanoid robots.

That means when the teams compete in the finals in Pomona Calif. next June for a $2 million prize, their robots won't just be asked to drive a car.They'll need to get out of the vehicle, too -- something that's much more complicated than it sounds.

Since driving is the first task the robots face, they won't be able to continue with the rest of the challenge if they can't manage that. Years of work will end with a quick failure.

DARPA, the Defense Advanced Research Projects Agency, will give teams an easy out: the option to walk the course, instead of drive and exit the vehicle. But any team that takes that route, won't be able to garner as many points as those taking on the driving and egress challenge.

And when it comes to beating the best robotics teams from around the world, the winning team is going to need all the points it can get.

For Worcester Polytechnic Institute, or WPI, that means tackling the hard stuff.

WPI robot Sharon Gaudin

WPI’s Atlas robot "Warner" autonomously reaches out and grabs a drill, a task it will need to master for the DARPA robotics challenge finals.

"It's a risky move, but if we're going to win, we've got to put all our money on the table and go full in," said Michael Gennert, director of robotics engineering at WPI. "We're not going to say, 'That's too hard.' We're going to do it. If we're going to win, we're going to win big. If we're going to fail, and I hope we don't, we're going to fail big, too."

DARPA's three-part challenge is intended to encourage the advancement of autonomous robots to the point that they could largely act on their own after a natural or man-made disaster, going into a damaged building, rescuing victims, turning off gas pipes and even putting out fires.

The first part of the challenge was a simulation held in 2013. The second part, which was took place in southern Florida last December, involved 16 teams competing to see which could build the best software to enable their robot to work through a series of individual tasks, such as walking, using tools and climbing a ladder.

During the June finals, the teams won't be facing individual tasks. Instead, their robots will confront a disaster situation that forces them to deal with tasks like removing debris, walking around or over obstacles, turning off valves or cutting into walls. If a robot can't complete a needed task, it won't be able to continue on.

Speed is another issue.

During the December challenge, the robots had 30 minutes for each specific task. Many failed to even open and walk through a door or climb over a small pile of debris in the given time. In the finals, they'll have just 45 minutes to an hour to accomplish all eight tasks.

"At this point, I'd say we're about 50% faster than last December, but we're hoping to get in the 75% or 80% range," said Matt DeDonato, the team's technical project manager. "It's a scary thing. It's daunting. With speed comes a lot of uncertainty and instability. As roboticists, we like everything slow because we can control slow. As you get more and more into the dynamic range, you have to make sure all your algorithms get updated so you can handle the higher speeds."

The WPI robotics team, which is working with researchers from Carnegie Mellon University, is already figuring out how to best get their 6-foot tall, 330-pound Boston Dynamics-built Atlas robot to maneuver out of a vehicle. (They've named it "Warner.") Of all of the known tasks they'll face -- DARPA has warned them that there will be a surprise one -- simply getting out of a car is the most daunting.

"The reason it's so hard is the robot is in contact with the vehicle at many points," Gennert said. "When it's walking, the robot touches the ground with its left foot and right foot and that's that. In a car, it has its fanny on its seat cushion, its back against the seat, its feet on the floor. It has its hands on the steering wheel. There are many and different kinds of contact. It has to shift its weight from the back of its legs onto its feet. That's really hard to do."

While the robot has sensors, it can't feel its legs or back pressing against the seat like a human does. Without feeling those points of contact, it has less information about its positioning, making decisions about its next move harder to make.

"Right now, we have one foot out and now we're shifting the weight onto that foot so it can move the other foot out," said DeDonato. "That's one thing we think will set us apart from the other teams. We were one of only two teams to actually finish driving the course [in the last challenge]. So we want to basically continue on that path."

The team, though, hasn't been spending all of its time on the driving task.

DeDonato said team members been working hard on the software needed to get Warner to pick up and use a drill, remove debris and walk over rough terrain more autonomously than before.

"We'll no longer be giving it joint-by-joint commands," he explained. "Last competition, it was a different level of autonomy. All the balancing was autonomous. When you told the hand to move, the robot didn't fall over. We gave it a lot of commands, like move to this point and reach out.... It was somewhat autonomous. Now we're giving it task goals. Walk there and pick up this object. He automatically figures out how to walk around stuff and grasp the object."

Copyright © 2014 IDG Communications, Inc.

It’s time to break the ChatGPT habit
Shop Tech Products at Amazon