DARPA Takes Robots to Disaster Mitigation Trials
By Cheryl Pellerin
American Forces Press Service
WASHINGTON, Dec. 12, 2013 Next week, 17 teams will take their multi-limbed, capable-looking robots through eight realistic disaster-response tasks that will make up the Defense Advanced Research Projects Agency’s Robotics Challenge Trials Dec. 21-22 at Florida’s Homestead-Miami Speedway.
Atlas is a high-mobility humanoid robot funded by the Defense Department and built by Boston Dynamics, an engineering company that began as a spinoff of the Massachusetts Institute of Technology. The robot is designed to negotiate rough outdoor terrain. It’s powered from an off-board electric power supply through a flexible tether, and its articulated sensor head includes stereo cameras and a laser range finder. Several copies of the robot are being provided as government-furnished equipment for the Defense Advanced Research Projects Agency’s Robotics Challenge program. Photo courtesy of the Defense Advanced Research Projects Agency
(Click photo for screen-resolution image);high-resolution image available.
The best performers will determine the baseline for the state of robotics, Dr. Gill Pratt, DARPA’s Robotics Challenge program manager, said during a recent teleconference. And DARPA will fund up to eight of the highest-scoring teams for another year as they move on to the DRC Finals in 2014, after which one team will receive a $2 million prize.
“The purpose of the program is to develop technology that can help make us much more robust to natural and manmade disasters,” Pratt explained.
“In particular,” he added, “we’re looking at robotic technology that can allow us to mitigate the extent of a disaster during the first hours and days while the disaster is still unfolding.”
DARPA was directly inspired to create the program by the 2011 accident at Japan’s Fukushima Daiichi nuclear power station, Pratt said, which was caused when an earthquake and tsunami knocked out backup power systems needed to cool the plant’s reactors, causing three of them to undergo fuel melting, hydrogen explosions and radioactive releases.
“During the first 24 hours there,” he said, “if only human beings had been able to go into the reactor buildings and vent built-up gas that was accumulating inside the reactors, the explosions that occurred might have been prevented and the disaster would not have been as severe.”
That’s just one example, Pratt added.
“We don’t know what the next disaster will be, so the technology we’re trying to develop [will] allow human beings and robots working together to have an effect on evolving disasters in environments that are too dangerous for human beings to go into by themselves,” he said.
DARPA is trying to improve robotic mobility and dexterity to achieve the following goals for disaster-response robots, Pratt said:
-- The robots have to work in environments that are engineered for people, including environments that are degraded by an evolving disaster;
-- The robots have to be able to use human tools, everything from screwdrivers to fire trucks that may be available in the disaster area; and
-- The robots must have an improved human-to-robot interface, to reduce the amount of training needed by personnel who are experts in handling disasters but not necessarily in handling robots.
“We started the program with over 100 teams and had a first event in June that was a virtual robotics challenge held in simulation,” Pratt said. Since then and through several design reviews, DARPA has narrowed the field to 17.
DARPA is funding 13 of the 17 teams, and four teams are funding their own work, the program manager said. Part of the funding includes a high-mobility humanoid robot called Atlas. It’s funded by the Defense Department and built by Boston Dynamics, an engineering company that began as a spinoff of the Massachusetts Institute of Technology.
The teams represent five countries and organizations that range from large and small businesses and hardware and software firms to universities and government agencies like NASA, which has two teams participating in the trials.
Each of the eight tasks the robots must perform has a couple of steps. The first task is to drive a utility vehicle over a short course that requires turning, then the robot must get out of the vehicle and walk, Pratt said. Second is to travel over rough terrain that goes from easy to medium to hard. Third is to move rubble from in front of a doorway and go through the door.
The fourth task is to walk through three successively more difficult-to-open doors. Fifth is to climb a ladder. Sixth is to go to a wall, pick up a tool and use it to cut an access hole through the wall without damaging infrastructure drawn on the wall. Seventh is to find three valves and close them. Eighth is to pull a fire hose a short distance and connect it to a standpipe.
The DARPA Robotics Challenge Trials are free and open to the public -- a public whose experience with robots may tend toward science fiction, Pratt worries, like the Terminator and R2D2, or lately even the Almost Human MX-263 combat-model android. And what will the public see next week at the Miami-Homestead Speedway?
Not all of the robots will be able to do every task, Pratt explained. Even those that can do most tasks will be getting a lot of help from their human operators. And the robots will be slow, he said.
“Right now, where we are is that robots are roughly at the same level of mobility and dexterity as a one-year-old child,” Pratt said, adding that each robot will have 30 minutes to do each of the eight tasks.
“What we’re doing with the DRC trials is we’re getting a calibration point,” he said. “We’re trying to understand the state of the art of the field.”
Today, Pratt said, real robots for the most part either work in on stationary bases in factories doing very clearly defined repetitive tasks or they are used in laboratories in schools in controlled environments. If robots are used outdoors they’re typically run through something called teleoperation, where a person dictates every move the robot makes each tenth of a second or more.
“We’re trying to advance that technology and move things from teleoperation to something known as task-level autonomy, where rather than ‘Move forward a tenth of an inch, move left a tenth of an inch,’ you tell the robot, ‘Open that door,’ and the robot perceives the handle on the door, reaches out, turns the handle and opens the door.”
Pratt said that’s the level of supervision he and others believe will be most effective for people and disaster-response robots to use to interact with each other.
Based on DARPA’s experience with its 2004 Grand Challenge for driverless vehicles, the program manager said robots that qualify for the 2014 DARPA Robotic Challenge Finals in 2014 should be much more capable than this year’s contenders.
“Let me paint a picture of where we hope we’ll get to,” Pratt said.
Take the eight tasks from this year’s trials -- going through doors, going up the ladder, moving rubble out of the way -- and imagine mixing the tasks into a single rather mission the robot must complete, he said.
“Let’s say we have a site that is a mockup of a disaster and … we give the robot a task: go rescue a person -- actually a dummy -- who’s hidden under a pile of rocks,” Pratt said. “To get to the pile of rocks there are ladders in the way, there are rubble fields, there are vehicles it can use.”
The desire is to physically emulate such a scenario roughly a year from now, and to have human beings in a remote location, able to control the robot over a degraded communication link, he said.
Pratt said DARPA is also focusing beyond search and rescue on operations that can help mediate disasters -- for example if there is a chemical leak in a factory and the chemicals are too corrosive for people to deal with.
“One possibility is putting people inside protective suits, but that only works for a very short time until oxygen runs out or it gets too hot,” he said. “A better idea is to separate the robot from the human being, have the person in a safe place and, despite having a bad communication level, allow the robot to do what a person in a suit would have done.”
Pratt added, “That’s our goal. How far we’ll get, we don’t know. Part of the purpose of the trials is to calibrate us as to where the field is now so we can design the finals to be a just-hard-enough test.”
(Follow Cheryl Pellerin on Twitter: @PellerinAFPS)