Separating science fact from science fiction in robotics (with video)

Many roboticists competing in the finals of the DARPA Robotics Challenge earlier this month, said they are backing away from autonomy in order to make their robots work faster and more efficiently.</p>

Science-fiction movies often show robots freely running across the screen, either wreaking havoc or saving the world.

Many roboticists competing in the finals of the DARPA Robotics Challenge earlier this month said they are backing away from autonomy in order to make their robots work faster and more efficiently.

For them, less autonomy for the robot and more human control means a better shot at winning the global, multi-year competition.

An autonomous robot is able to complete tasks or behaviors without human intervention. The robots competing in the DARPA challenge were semi-autonomous, meaning they performed some tasks on their own but needed human guidance for others.

"We still have a long way to go with autonomy," Dennis Hong, leader of the UCLA team in the DARPA challenge, told Computerworld. "Originally, when DARPA launched this competition, autonomy was a huge part of it ... but many teams aren't using much autonomy [in the finals]. Autonomy is hard and they don't have the time and money to build autonomy."

Hong, whose team came in 13th out of 24 teams at the finals, said before the final competition that his team would use more human-controlled operations than autonomy.

Robots run with human operators are still faster. "This challenge pushed state-of-the-art in locomotion and humanoid robots, in general," he said. "But not really in autonomy."

Scientists are not even close to building a robot that can do laundry or make dinner. Robots today are also not at the point where they can rush into burning buildings, shut off systems and rescue victims, although that's where DARPA sees the technology going in another 10 or 15 years.

That being said, robotics, and autonomy, are far more advanced than they were just a year and a half ago.

At the Robotics Challenge Trials in December 2013, the machines were tethered to their human controllers, who directed the robots on such a granular level that they had to instruct them to move a hand ahead 2.5 inches, move the shoulder, turn the torso and then turn the wrist to a certain point and close the fingers around a door knob.

Today, most of the scientists at the robotics finals could use computerized commands via Wi-Fi to make the robot walk to a door and open it.

The robots are able to generally balance on their own, walk forward, grasp tools, and turn valves and knobs.

In terms of autonomy, it's a huge advance.

It's just not the advance that people, who get most of their perceptions about robotics from movies and television shows, have been expecting.

"Our team's focus was to not maximize autonomy but to maximize the utility of having a human-robot team," said Taskin Padir, assistant professor of electrical and computer engineering at Worcester Polytechnic Institute, which, working with Carnegie Mellon University, placed seventh in the DARPA challenge. "Autonomy is still hard. It takes time if it's easier for the human to help the robot with tasks, we do that."

The WPI team developed an algorithm that could enable a robot to recognize a door handle in 10 seconds, Padir said. As exciting as that was, a human could do the same thing in a moment, so the team used the operator, instead of autonomy, to get the robot to perform the task.

"At the end of the day, the success was in how we maximize the efficiencies of using humans to help make the robot act more efficiently," Padir said. "None of it was saying, 'Robot, go turn the valve. Robot, go cut the wall.' None of it was fully autonomous."

In the weeks leading up to the finals, WPI's team focused on using more autonomy to make its robot faster.

The team needed the help. A month before the finals, WPI's robot, nicknamed Warner, still needed two hours to make it through the DARPA course of eight tasks. Those tasks included driving a car, opening a door, climbing stairs and maneuvering over rough terrain.

The problem was that they would only have a single hour to complete those tasks during the competition.

At the time, Matt DeDonato, the WPI team's technical project manager, said the team tried to make the human operators running its robot more efficient, while also increasing the robot's autonomy so it could do more work without waiting for its operators to send it guidance.

Felipe Polido, a senior robotics engineer on WPI's team, said, in the end, the team used a lot more autonomy with the robot than it had during the 2013 trials, but relied heavily on a combination of human and robot teamwork.

"Before, we would tell it to move its arm this much, move its shoulder this much,"Polido said. "This time, there was much more autonomy because the robot could just move its arm on its own. For the plug task, the robot moved its hand to the plug but the operator helped it grab the plug."

The team from NASA's Jet Propulsion Laboratory,however, used less autonomy in the finals than it did during the trials.

"We shifted more to the operator to make it faster," said Brett Kennedy, principle investigator for JPL's team in the robotics challenge. "The challenge isn't really about autonomy but about getting the job done. We're not going to force autonomy on the situation."

Pam Melroy, deputy director of the tactical technology office at DARPA, said the fact that different teams used different levels of autonomy and that humans are still trusted to be faster than robots is part of what DARPA officials wanted to learn at the challenge.

"One of the purposes of the DARPA robotics challenge is to show science fact vs. science fiction," said Melroy, who is a retired U.S. Air Force officer and a former NASA astronaut. "State of the art in autonomy is extremely challenging. We're just not as far along as people think we are, especially when you mix in the software and hardware together in a real-world environment."

She noted that what many teams learned was the efficiency of bundling several tasks together. If a robot is doing two to four tasks in a row, then the human operator has time to plan for the next steps.

"Even two or three things programmed in make a big difference. You can see it," said Melroy. "It's baby steps."

Join the TechWorld newsletter!

Error: Please check your email address.

Tags OriginAutonomyroboticscomputerworldDARPA

More about AutonomyMellonNASAUCLA

Show Comments
[]