DARPA’s Humanoid Robots Take a Slow-Motion Leap Forward

If you don’t know DARPA, you should.

The Defense Advanced Research Projects Agency is a division of the military charged with coming up with the coolest, most useful possible new technologies for our defense forces. It’s a U.S. government agency that routinely fires on all cylinders, backs good ideas, and improves the lives of us all.

The great thing is that DARPA frequently winds up producing rich fruit for the entire world to use; its past projects have spawned such hits as Google’s self-driving cars, GPS, Siri, and, by the way, the Internet. (Here’s a really entertaining interview I conducted with Dan Kaufman, one of DARPA’s directors.)

Now, DARPA is tackling robots. Not mechanical arms in car factories and the like—real robots, the kind we’ve seen in movies for decades, the kind little kids picture when they hear the term “robots.” We’re talking about humanoid, two-legged, walking, thinking machines.

Why? In 2011, an earthquake and tsunami in Japan triggered a meltdown at the Fukishima nuclear power plant. Workers fled the plant before they could shut it down. And that got DARPA thinking: If only we had robots that could go in and finish the job, we could save a lot of lives.

So DARPA did what DARPA does best: It offered a Challenge, open to anyone daring enough to enter — universities, companies, or individuals. It offered $3.5 million in prizes to teams who could successfully complete eight tasks in a simulated power-plant rescue scenario:

1. Drive a vehicle to the plant, stop and park;

2. Get out of the car;

3. Walk to the plant’s doorway, open its human-sized handle, and walk through;

4. Turn a valve handle one full rotation;

5. Pick up a power drill, turn it on, and cut a hole out of a sheetrock wall big enough for a person to escape;

6. Climb over (or push aside) a room full of rubble;

7. Exit by climbing a metal staircase; and

8. A surprise task.

The surprise changed on each day. On Day One, robots were asked to pull down a lever in a fuse box; on Day Two, they had to pull a power cord out of a socket and plug it into a neighboring socket.

The fun began in 2013, at the DARPA Robotics Challenge Trials. At that competition, the robots had wires coming out of them to provide power and allow them to communicate with their human operators, who were driving the robots by remote control. The robots had harnesses, so there was no risk of falling over and smashing. They had 30 minutes to complete each of the eight tasks.

The New Rules

And then last weekend—only eighteen months later—DARPA expected the robots to tackle the Robotics Challenge Finals with much tougher rules. (Yahoo Tech was there; see the video above.)

This time around, there were no power cords, no wires, and no harnesses. The total time allowed to complete the course was one hour instead of four hours.

And toughest of all: Once the robot entered the “plant,” DARPA would deliberately degrade the operators’ wireless signal to the robots by introducing communication blackouts up to 30 seconds long. According to the rule book, “the blackouts will be structured so that teams with more autonomous systems will be able to progress through a run more quickly.” In other words, they wanted to find out how well the robots thought for themselves.

The Competition

DARPA selected 25 teams to compete in the finals, including entries from Japan, Korea, China, Hong Kong, Germany, South Korea and Italy. Most were humanoid. Some looked like centaurs; some, like Carnegie Mellon’s entry shown below, had wheels on their “elbows” so that they could convert into rolling vehicles.

The robots moved very slowly; often, they’d just stand there, motionless, as they computed their next moves or created images of the scenes before them. But team MIT’s leader, Russ Tendrake, told me that speed wasn’t the goal here. “We have to be conservative and slow, and make sure that we get the job done. If we were putting on a show, if we were just trying to go fast and show people how fast the robots could move, we could have them walk about eight times faster than we’re gonna do today.”

So to prevent the thousands of spectators from getting bored, DARPA cleverly built four identical power-plant sets, side-by-side. Most of the time, something was happening on one of the courses. Five Jumbotron screens let everyone see what was going on during the two days of the competition.

A huge exhibit area was set up outside the arena, where dozens of robotics companies displayed their wares—and let showgoers try their hand at operating real robots.

Pathetic or Amazing?

Maybe you’ve seen the viral video montage of robots from the DARPA finals falling over. They did that a lot.

Even MIT’s robot fell as it was getting out of its car. “We made a simple operator mistake,” Tendrake says. “When we were working through our checklist, we forgot to tell the robot it was done driving, and that it should start balancing. Basically, it tried to drive while it was getting out of the car. Our big, lumbering, 370-pound robot face-planted at full speed into the ground.” (It lost an arm in the fall, but heroically completed the course one-handed, finishing with a respectable 7 out of 8 points.)

The surprisingly big, enthusiastic crowd anthropomorphized these machines like crazy.

“It is an extraordinary thing, isn’t it?” said DARPA’s Gill Pratt, who ran the Robotics Challenge. “We had not expected it. When the robot does well and it scores a point, everyone cheers as if they’re the ones getting the point. And then, of course, when the robot teeters and then suddenly falls, everybody goes, ‘Awwwww!’, and they sympathize with it. I had a person last night ask me whether a particular robot that had fallen and gotten damaged, whether it was ‘still hurt!’ This is a machine. This is aluminum and steel and some copper wiring!”

And so, on one hand, these state-of-the-art machines did a pretty good job of reassuring us that we’re in no immediate danger of being threatened by rebellious robots (see: Terminator, I, Robot, 2001, and so on).

On the other hand, 10 years ago, humanoid, two-legged robots didn’t exist at all. The pace of improvement, even since the trials in 2013, is stunning. And this technology is still in its infancy.

“These are still the Model Ts,” Pratt told me. “In coming years, reliability will go up, prices will go down. And the technology here can be applied to robots for many, many purposes— in particular, robots for an aging society. In the United States right now, 13 percent of the population is over age 65. In 15 years that number will rise to 20 percent. Who’s gonna take care of us? Wouldn’t it be better if I could age in place, if I wouldn’t have to go to a nursing home? If there would be technology in my home to make life better, so that I can have a more dignified life as I grow old?”

All technologies start off crude and slow. Computers, the Internet, cellphones, you name it. Watching these robots struggle, fall, and get up again—sometimes on their own—it was easy to imagine that after only a few more generations of improvement, these will be truly useful robots in our homes (and on our battlefields).

In the end, the $2 million first-place winner was HUBO, a robot from a South Korean team called KAIST. It performed all eight tasks flawlessly, in 44 minutes. The $1 million second prize went to the Institute of Human and Machine Cognition in Florida. Carnegie Mellon landed the $500,000 third prize.

And now, as of this week, the two-year, multimillion-dollar DARPA Robotics Challenge is over. Having planted the seeds of the robot revolution, DARPA now intends to do what it always does: Step back and wait to see what blooms.

David Pogue is the founder of Yahoo Tech. On the Web, he’s davidpogue.com. On Twitter, he’s @pogue. On email, he’s poguester@yahoo.com. He welcomes non-toxic comments in the Comments below.