It's now or never to speed up robots for DARPA challenge

WPI robotics team has just 3 weeks to make its robot operate twice as fast

WORCESTER, Mass. -- The robotics team at Worcester Polytechnic Institute has three weeks before the finals of the DARPA Robotics Challenge to make their robot twice as fast as it is today.

"We need two hours now to get through the course," said Matt DeDonato, the WPI team's technical project manager. "If we had two hours, we'd be golden, but we're only going to have one. So we need to speed it up."

The WPI team is one of 25 teams qualified to compete in the robotics challenge finals on June 5 and 6 in Pomona, Calif. The challenge, which launched in 2012, is intended to encourage roboticists to build robots that can one day provide support in a disaster.

In the last challenge, held in December 2013, each team's robot was required to perform eight tasks, including climbing a ladder, driving a car , opening doors and using a drill. The robot had 30 minutes to perform each task.

This year, the robots will face a course that simulates a disaster situation and will have to take on each task one after another. The robot must complete all the tasks in one hour.

DeDonato said WPI's humanoid robot is capable of completing each of the different tasks. The issue is speed.

The WPI team had planned to use the cloud to speed up their robot during the final competition, but the team was forced to change plans.

Michael Gennert, director of robotics engineering at WPI, said a team of students is working on putting software commands in the cloud so the WPI robot – named Warner – could access those instructions anytime and get directions even if the Wi-Fi connection with the robot's operators was down. That would have meant that even if the robot was working without its controllers, it still could access information and pre-set directions.

However, the WPI team was preparing to use Amazon's cloud platform and only recently found out that DARPA will only give competitors access to Microsoft's Azure platform. The work they had done for the Amazon platform is not transferrable to Microsoft's cloud.

Warner, WPI's humanoid Atlas robot Sharon Gaudin/Computerworld

Warner, WPI's humanoid Atlas robot, gets ready to practice walking over a pile of debris, which will be one of the tasks it has to take on in the finals of the DARPA Robotics Challenge.

"It's a little disappointing, but we have to show that the cloud can improve robotic performance," Gennert said. "In the bigger picture, we'll certainly see more use of the cloud and that's one of the things that will help us double robotic performance in 18 to 24 months."

With the cloud no longer an option, the WPI team is tweaking its algorithms to try to speed up Warner, but the group is also working to make the robot's human operators work faster and more efficiently.

In the 2013 competition, the operators gave the robot commands for most of what it did -- how far to turn its wrist, how far to extend its arm or how many steps to take in which direction. Today's robot is much more autonomous.

Operators will tell the robot to open a door but no longer need to tell the machine how to position its body, such as at the shoulder and wrist. The robot can make those calculations on its own and more quickly than an operator could instruct it.

That makes the robot operate much faster now than it did a year and a half ago. However, it's still not fast enough, so DeDonato will be working with the robot's operators to make improvements. There's a separate human operator for each task that the robot needs to perform. At the end of each task, the operator needs to make sure the robot's body or arms are in the correct position so it's ready to quickly move on to the next task.

"It's not that complicated, but you have to worry about everyone who's driving," said DeDonato. "When does one task finish and another start? Where is that tradeoff made? Is the driver leaving the robot in the right state to start the next task?"

WPI's 7-foot-tall Atlas robot Sharon Gaudin

WPI's 7-foot-tall Atlas robot opens and walks -- sideways, to allow for its size -- through a doorway on a mockup course that mimics what the robot will have to take on during the finals of the DARPA robotics challenge.

The WPI roboticists also discovered that during the task in which the robot moves through debris, such as broken boards and two-by-fours, it's easier and quicker to have the robot shuffle through the debris, pushing it aside with its feet, instead of stopping to bend over, pick up a piece of debris and move it.

Some tasks are easier than others. Warner, for instance, can turn a valve, open a door and walk through it and climb stairs fairly easily and quickly.

The drill task is different. It is the hardest and most time-consuming job for WPI's robot. The task doesn't involve one step. It includes finding the drill, picking it up, turning it on and then using it. Making it more difficult, the team isn't sure what kind of drill will be used during the competition.

"We can do it. It just does take some time," said DeDonato. If the team can't speed up the robot on the drill task, it might have to skip doing it, and lose the point for accomplishing the task. Skipping a task that takes a long time might enable the robot to finish the course and the rest of the tasks, giving the team a chance at more total points.

While the team is trying to build up the robot's speed at basic tasks, Gennert said the group is moving much faster than it did a year ago. One day soon, with more research, robots will move far more easily and quickly.

"Ten years ago, it would have taken hours for the robot to do these things, if at all," added Gennert. "In 10 or 12 years, they could be as fast as humans at these tasks. What can be accomplished has increased rapidly."

To express your thoughts on Computerworld content, visit Computerworld's Facebook page, LinkedIn page and Twitter stream.
Windows 10 annoyances and solutions
Shop Tech Products at Amazon
Notice to our Readers
We're now using social media to take your comments and feedback. Learn more about this here.