HOMESTEAD, Fla. — The robot gently grabbed the door handle and pulled it open. But before it could shuffle through the frame, it lost its grip and the spring-loaded door slammed shut.
Team Schaft, an elite group of former Tokyo University roboticists whose company was recently acquired by Google, is one of the favorites to win the Pentagon’s Darpa Robotics Challenge 2013 Trials. But in its first event on Friday, the team’s 4-foot-8, 210-pound Schaft HRP-2 robot was not perfect. The robot passed through just two of the three doors and as a result failed to get a valuable third point or a fourth bonus point.
With computer technology and robot development rapidly advancing, the Robotics Challenge is a coming-out party of sorts for a new generation of robots that are being designed to walk around in the human environment and collaborate with people on tasks. But the event is much more than a robot showcase; it is also an attempt by the federal government to push technology forward, with the goal of creating machines that can be used in place of humans to perform dangerous tasks.
Nine years ago, Darpa — the Defense Advanced Research Projects Agency — sponsored the first of three autonomous vehicle “grand challenges.” That has led to an explosion of activity in self-driving cars.
The agency is trying to repeat its success by offering a $2 million prize next year to the team that develops a robot that can perform a series of tasks simulating the kind of things a robot might be called upon to do in a disaster like a nuclear power plant meltdown.
There have been a series of prechallenge events this year, building up to the “trials” here on Friday and Saturday. Even with all of that, expectations have been set low.
Rightly so, apparently. The Schaft robot, even with its imperfections, performed better than many of the other competitors who have struggled with far easier tasks, like pressing the gas pedal in the golf-cart-like vehicle that the robots are supposed to control or making their way up a ladder. Many of the competitors are not even trying all of the eight required tasks, which include driving a car, breaking through a wall, operating power tools and removing debris.
Gill Pratt, the roboticist and Darpa program manager who is in charge of the event, said Thursday that public perceptions of the state of robot science were unrealistic. “We’ve all seen science fiction movies,” he said at a news briefing. “And in science fiction movies, robots walk around quickly, they do nice things, sometimes not so nice — and in general they’re inside of action films.”
He said one reason for the Darpa trials was to establish where the science stood, “as a reference point, as a way for us to learn what is the state of the art right now.”
By the look of things here in the pit lane of the Homestead-Miami Speedway, humanoid robots will remain more science fiction than science fact for a while longer.
Dr. Pratt, who before coming to Darpa was a pioneer in developing technologies that made it possible for robots to work in proximity to humans, has drawn an analogy between today’s robots and human 1-year-olds, who can barely walk and frequently lose their balance and topple over.
The question is how quickly the robots will progress. In the case of self-driven vehicles, the rate of progress was much faster than predicted.
Although the original event ended in failure in 2004, just a year and a half later a number of the vehicles were able on their own to complete a long course over desert roads. In 2010, Google shook up the automotive industry with the news that it was testing self-driving cars on California’s freeways. Now, in the 2014 model year, a number of car companies will offer a feature known as “traffic jam assist” in which cars will drive autonomously in limited freeway situations.
To add realism to the 2013 Challenge event, Darpa officials have made it more difficult to control the robots by varying the amount of Internet bandwidth that is available to operators at its computer control station. In theory, that rewards teams whose designs are built around more self-guidance and punishes those who manually control or “teleoperate” their machines.
At the trials, some autonomy is on display. For example, the Atlas robot that was designed for seven of the teams by Boston Dynamics, a Massachusetts company, has the ability to walk on its own, as well as balance, a challenging robotics feat. The Schaft robot climbs over a debris field largely on its own, with the operator selecting the positions for the robot to place its feet, but the robot then taking the steps on its own.
Although the trials have shown so far that the teams are struggling to reach even the low bar that Darpa has set, there are expectations that a year from now, considerable progress will have been made. Indeed, at least one of the teams may try to automate their performance entirely next year.
Darpa has specified, down to the tiniest detail, what it wants from the teams in the way of the contest, which will help any teams that want to go back to the lab to try again.
“It’s not that different from the challenge faced by Disney engineers,” said Aaron Bobick, a Georgia Tech computer-vision specialist who is a member of the DRC-Hubo team led by Drexel University. And like the Walt Disney Company when designing a new thrill ride, he said, Darpa participants find that the challenge is to keep the ride on the rails.
Source : The New York Times