ARPA Robots to Face Final Challenge in California

By Cheryl Pellerin

DoD News, Defense Media Activity

WASHINGTON, July 15, 2014 – The Defense Advanced Research Projects Agency’s third and final challenge among 24 or so U.S. and international human-robot teams will take place in California next June, ending with a $2 million prize and robots that for the first time may be capable of helping first responders save lives when a disaster strikes anywhere in the world.

The main goal of the DARPA Robotics Challenge program is to develop ground-robotics capabilities for executing complex tasks in the dangerous, degraded human-engineered environments created when disasters strike cities.

“The purpose is to protect lives during manmade and natural disasters,” DARPA program manager Dr. Gill Pratt told reporters during a recent media call. The program began in 2012, but DARPA has been trying to use robots to help in disasters since 2001.

In the days after 9/11, DARPA sent to New York City robots whose development the agency had funded. But those robots found no survivors, Pratt recalled in an analytic piece published last Dec. 3 in The Bulletin of Atomic Scientists.

DARPA officials tried again in March 2011 when a magnitude 9.0 earthquake centered off the coast of Sendai on the eastern coast of Honshu Island, Japan, produced a 49-foot tsunami that killed 19,000 people, destroyed a million buildings and flooded Tokyo Electric Power Co.'s Fukushima Daiichi nuclear plant.

In the plant, the reactor cores of three operating units melted, and a fourth was damaged. Japanese officials declared a nuclear emergency and ultimately evacuated people within 12 miles of the plant.

Humanitarian assistance and disaster relief is a primary DOD mission, Pratt wrote in the Dec. 3 Bulletin, and as the disaster unfolded in Japan, “DARPA officials contacted researchers who had designed robots for the Three Mile Island and Chernobyl [nuclear] disasters and coordinated with companies that DARPA had funded to develop other robots.”

Each company already was making plans to send its robots and training personnel to Japan, he added, and others around the world sent robots, but it took weeks for power-plant personnel to complete the training they needed to operate the robots.

By then, Pratt said, it was too late for the robots to help.

“A key idea here is that these robots don't operate on their own,” Pratt said during the media call. “In fact, the state of the art is not capable of having a robot do useful work on its own in these very difficult environments. So we partner them with operators who supervise the robots … at a distance from the disaster zone, connected through a communication link to the robot in the disaster zone.”

In such a team, he added, “a robot does what it's best at, which is surviving difficult conditions in the disaster, and the human being does what they're best at, which is using human perception, planning and experience to tell the robot what to do.”

The DARPA Robotics Challenge launched in October 2012 and held two competitions in 2013 -– a virtual event in June and a two-day event in December at the Homestead-Miami Speedway in Florida.

The first competition tested software teams’ abilities to guide a simulated robot through three sample tasks in a virtual environment. In December, teams had to guide real robots through as many as eight individual physical tasks that tested robot mobility, manipulation, dexterity, perception and operator-control mechanisms.

At the trials in Miami, Pratt said, “we started with 16 teams and … went through eight different tasks, from cutting a hole in a wall using a tool, climbing a ladder and traveling over rough terrain, and even driving a small vehicle that a robot might be called on [to use] to go back and forth between [a safe area] and a disaster zone.”

DARPA officials developed the tasks in consultation with the teams, other experts and first responders, the program manager explained, adding that DARPA is not trying to match team skills to a particular kind of disaster.

“We try to use inspiration from one disaster, like Fukushima or the ferry disaster in [South] Korea, and abstract away and talk to first responders -- we've done that quite a bit now -- and say, ‘What's the common thread?’”

Bad communications almost always are a common thread, Pratt said, along with large areas of rubble and debris. First responders describe what bad comms or debris are like in a disaster zone, and DARPA comes up with a model for the robots.

“Often, what happens is that what we come up with is too hard for the robots,” Pratt said. “So if you look at the trials, you say, ‘Did the rubble in the trials disaster look like rubble in Fukushima?’ And the answer is, ‘Not even close.’ But we have to get there, and this is the slope we're trying to climb in terms of difficulty.”

But something did happen at the trials in Miami that no one, not even Pratt, expected.

“It turned out that things went better than we expected,” he said, adding that the robots were more reliable than expected, with better mobility, grasping and manipulation ability.

Because of that success and other factors, he said, DARPA officials are changing the scope of the program to raise the bar at the finals even more than they had planned.

Pratt said the other factors include a significant upswing in commercial investment in robotics, decisions by the governments of Japan, South Korea and countries in the European Union to sponsor and fund teams to participate in the finals, and a new concept in robotic autonomy called “cloud robotics.”

In cloud robotics, he explained, the robot is able to exploit remote information and remote computing capability on the Internet through a high-speed link and share information to increase their effectiveness by reusing information that's provided from past sources.

“We think that particular technical advance has a lot of promise, and we believe the commercial world is going to take off with it,” Pratt said. “But we want to exploit cloud robotics and the investment that's coming from other parts of the world in a way that is applicable to disaster response.”

That means doing work DARPA officials believe the commercial sector will not do, the program manager said -- “in particular, problems that are unique to disaster response.”

One of these is operation without the possibility of physical human intervention if something goes wrong.

“In a disaster, the reason you use a robot in the first place is because the environment is very harsh, and you can't send a person in,” he explained. “So we have to make sure the robot will continue to work well even if there's no way a human being can physically go there to help out.”

Such an environment will be more austere than it is in a home environment or a factory, or even on a farm, he added, so the robots must be more capable in locomotion and manipulation than under normal circumstances. Maybe most importantly, the connection to the cloud will be intermittent, Pratt said.

“In disasters typically communications … suffer most, so we are going to purposefully try to emulate the very degraded communication environments that happen in real disasters,” he said. “We don't think that's something the commercial world will try to tackle in the near term.”

But for the robots’ human supervisors during the finals, DARPA will provide high-bandwidth links that go between the operators and their computers and the Internet, and teams will be able to use as much cloud computing power and computer disk storage, and also may use as many other experts as they like to help them help their robots.

To accommodate such evolutionary changes in the program, DARPA has added six months to the original timeline for the finals – moving from December 2014 out to June 2015.

Total funding for the DARPA program, from October 2012 to June 2015, is $95 million. DARPA-funded teams will receive $1.5 million between now and June -- other teams are self-funded -- and the team that wins will receive $2 million.

Tasks for the finals are not yet solidified, but Pratt said they will be similar to tasks in the Miami trials, with some modifications.

“Instead of being eight separate tasks, each one of them done pretty slowly, we're going to put all the tasks together into a sequence that is much more authentic to a real disaster,” he said. “For instance, you have to drive the vehicle to the site, get out of the vehicle, climb up the stairs, go over the rough ground, and each one follows the next, and the robot doesn't have a choice,” he added. “It must keep managing to make it through the next challenge, and each one happens … right after the other one.”

Other differences include the following, Pratt said:

-- The robots will not be connected to any kind of physical tethers or wires. Communication will be wireless, power sources will be onboard the robot and must allow the robot to run for one hour, and the human supervisor won’t be allowed to physically intervene. “If a robot falls or gets stuck, the fall will have to occur without breaking something on the robot that is vital for its continued operation,” Pratt said, “and the robot will have to be able to get up without assistance.”

-- All eight tasks must be completed in less than an hour, meaning that robots in the finals will be asked to go at least four times faster than they did at the Miami trials.

-- Communications will be degraded to a greater degree than they were during the Miami trials, to be more authentic to real disasters. “We think it's going to require quite a bit of innovation from the teams to adapt to our adjustment of the goal,” Pratt said. “We're sort of raising the bar, so … we're going to give them more time and more funding to get that done,” Pratt said.

-- One of the tasks will be a surprise to all teams.

In general, Pratt said, “we’ll give teams less prior information as to the specifics of the tasks. We're trying to slowly move things so that we're closer to a more authentic test of what a real disaster would be like.”