TechRepublic takes you inside the strange world of the 2015 DARPA Robotics Challenge, filled with running, swimming, falling, skating, beeping robots — and some brilliant humans.
Here in the middle of the desert, the sun is blazing and no Wi-Fi hotspots are allowed. Hundreds of government-affiliated organizations stand by giant tents filled with electronics, virtual reality headsets, pieces and limbs of bots. One booth is marked 'unmanned aerial vehicles," another simulates emergency disasters with first responders. A robot cheetah crouches in a glass case, silently watching.
Buzzing, whizzing, bumping, life-size R2D2s beeping and whistling. Wheels scrape over asphalt, feet sweep cords out of the way. A young girl tries on 3D printed robot hand. She picks up blocks and robot-high-fives her friend. Laughs. Sighs. Faint scents of greasy pizza, popcorn, cheap coffee hang in the air. Cargo shorts and tennis shoes pound the Earth. It's a sea of matching polos with white embroidered logos out here. Solidarity.
Something's head falls off, rolls out of the tent. No big, someone picks it up and brushes it off, eyes scanning the crowd. Fingers type code. Fingers fix code. Guy eats Ritz crackers. The programmers, they're hunkered down, backs bent, eyes glazed over. Listen, closely, to the company schpiel by the public affairs team, rotating promo videos. Meanwhile, a robot walks on a treadmill, looking bored. Next door, a smaller version is suspended mid-air, dangling from a noose around its non-existent head. Three aisles down, another robot seems to be sleeping, while its friend does a jig.
All of us, waiting, waiting, for the eyes to come alive. They don't. But the humans will keep trying. They're competing. They want to be first. They want to be the ones who get the closest.
RUN DRC states a t-shirt. Good Robot says another. I love robots, kids yell.
"There is no final frontier," says DARPA.
The end of the challenge
An old horse racing track stands in Pomona, California, surrounded by breathtaking mountains and sky-high palm trees. From June 5-6, 24 teams from around the world gathered here at the Fairplex for the DARPA Robotics Challenge (DRC), a competition funded by the US Defense Advanced Research Project Agency (DARPA) that has been running in three phases since 2012.
It's ironic, because watching a robot is a lot like watching a horse race. There's mostly silence and bated breath. There could even be some betting, who knows. But every few minutes, when something happens, the audience erupts in applause — for a step, a crouch, a raised arm, a turn. The slightest movement yields an incredible response because this a rare chance for the general public to see semi-autonomous humanoid robots in action.
These robots aren't just for show. This year's competition started after Fukushima, to develop robots capable of assisting humans in responding to natural and man-made disasters.
Pomona was chosen because of its miniscule chance of rain. But it did rain on Thursday in Los Angeles — just a sprinkling, a seemingly one in a million chance for Southern California, which is experiencing the worst drought in its history.
Luckily, the sprinkling didn't hit in Pomona. Robots don't like getting wet. Well, except those unmanned swimming underwater ones. Those aren't the focus here, though. They're just a parking lot sideshow.
Back in the grandstands, hundreds of people gathered to watch a trial run the night before the competition. This robot was having a tough time. It arrived at the site driving a Polaris, opened the door, walked through. One step, two. Three. Four. Stops, turns, stops, turns. We're all waiting. A minute passes by.
"A lot of pregnant pauses, aren't there?" someone says.
Something could be going wrong. The sensors could be malfunctioning, the motor might not be working correctly. Could be the software. The communication lines are degraded while the robot is inside the makeshift building, filled with debris from the "disaster."
"Could be a human operator that's drunk," someone jokes. "You never know."
Probably not the case, although the teams seemed delirious from exhaustion as Friday afternoon, the end of the first day of the finals, neared. On the course, behind a bright orange barricade, the hardware teams carefully watched the robot. Thousands of yards away, in Garage Nine, the software team hunched over computers. They spent weeks programming the robot to do various things — in particular, manipulation, so that it could autonomously complete small tasks like turning knobs or stepping over rubble.
The robots have to complete eight tasks in the finals. First, they must drive a utility vehicle (a Polaris Ranger) down a dirt road, with two small barricades to avoid. They must exit the vehicle, approach a door and open it. This is when the communication starts degrading. Periodic blackouts occur while the robot is in the "building."
It must locate and close a valve by turning it, then pick up a drill and cut through the wall. There is then a surprise task, different each day. On Saturday, teams had to pull one cord out of the wall and plug it into another hole. After that, the robots must either climb over broken cement blocks or move them out of the way to push through, exit through a doorway, and climb stairs.
A person can complete the tasks in five to 10 minutes. The robots had one hour, and were docked minutes if humans had any interaction with them if they fell or stopped working.
By the end of the day on Friday, Team Tartan Rescue, out of Carnegie Mellon University, had eight of eight points with their robot CHIMP, which completed the challenge in 55:15.
Team NimbRo Rescue out of Germany had seven points in 34 minutes, RoboSimian out of NASA's Jet Propulsion Laboratory had seven in 48 minutes, and MIT's Atlas robot had seven points in just over 50 minutes. Overall, six teams had seven points, one had six, and the rest had five or less.
Almost all the robots were bipedal, except for JPL's RoboSimian and the German NimbRo. Atlas, the humanoid robot developed by Boston Dynamics and MIT, was expected to show off — was used by seven teams and is endorsed by DARPA. It did well, but the four-legged ones excelled.
Outside the track, RoboSimian's backup — or organ donor, as Jason Reid from JPL called it — was on display at the expo. Each leg has seven degrees of motion, so the software team had to account for 28 possible movements wherever it goes.
That's quite a bit of programming. The team hasn't had a day off in months.
A tale of two robots
Garage Nine is where 24 teams have been living and working for a week, their booths filled with mock obstacles, bags of gravel and soil, bowls of Dorito crumbs and cups once filled with Coca-Cola, blankets and pillows, printed out code. The energy is high, all the time. They're swapping tools, giving advice, meeting up with former colleagues and classmates. They were also juggling, but they had to stop. Didn't want to harm the robots.
The robot spectrum is wide. Some are funded by DARPA, or other government organizations or businesses, or universities. But all the teams made it through the trials, no matter their age or experience level. They're all here, in the garage, competing for $3.5 million in prizes.
It's mostly men. About 95% of the teams are comprised of males. Eleven out of the 24 teams were all male. In total, only 23 of the 444 members are women. DARPA — and robotics programs worldwide — are facing the same gender issues that the entire tech industry is facing. And here in Pomona, it's a glaringly obvious problem.
The busiest booth was Tartan Rescue, out of Carnegie Mellon University and the National Robotics Engineering Center. CHIMP, as it's called, built by a team of professional engineers and research sponsors like FoxConn and Amazon, and is constantly surrounded by awestruck people. Tony Stentz, the team lead, watched as CHIMP went through a test run Saturday morning. On Friday, everything that could've went wrong, did. It got stuck in the car. It fell. It needed a joint replaced. But it still made first place. The team of young men fixed things quickly. They're well-disciplined, Stentz said.
At the far end of the garage is University of Hong Kong, with its Atlas robot. Atlas didn't even make it through the door of the building on Saturday, so they only got two points. Having the Boston Dynamics team next door to ask questions about Atlas has been an advantage for Team HKU, Tommy Hu said. He's one of the team members, an undergraduate student. Most of the HKU team are students, and this competition is a huge learning experience.
Atlas fell for the first time the night before the first run, on the concrete garage floor. They picked it back up, tested its joints. They pushed its limits, probably too far. But Saturday is the big day, so falling is sure to occur again.
"It's a lot less stressful now, emotionally," Hu laughed.
The not-so-final frontier
First thing on Saturday morning, a robot crashes. It echoes throughout the track. A collective gasp from the audience. Teammates rush over to pick it up. That's their pal. Is he hurt? Nah. Dust him off, set him back up. Everyone applauds, as if it's a basketball star who limped off the court with the utmost dignity.
Save for the falls and the stair stepping, most of the competition was just watching robots stare at walls.
"It's not quite like watching paint dry... think of it like a golf game," said Gill Pratt, program manager for DRC.
Preparing people for that is important, and it was obvious that although people were impressed by the robots at DRC, most people were surprised by how little occurred. And that's because the majority of the general public's robotics knowledge stems from Isaac Asimov books, Will Smith movies, and Terminator video clips.
We expect them to run and jump, their eyes to light up as if they understand every word we say. The fact is, they're dumb. Really dumb. But being able to complete these small tasks is huge improvement, and why DARPA focused DRC on semi-autonomous manipulation.
Picking up a drill, or turning a valve, are some of the hardest things to program, said Richard Mahoney, of SRI International
"We need to find ways to communicate what robots can do and how they can be helpful," he said.
Manipulation is incredibly important for the future of robotics, but it also takes the longest to figure out. SRI leads projects for DARPA and other government organizations to do that. They're working on building a robot miner, controlled by humans, and robotic suits for soldiers to carry heavy loads or to help children with muscular dystrophy. The humanoid robot SRI unveiled at DRC lasts eight hours on battery (the Atlas lasts only an hour). It's partially 3D printed and moves more like a human.
Mahoney said he thinks it will be 10 to 20 years before we see a humanoid robot in action — that is, more action than dangling from a cord and walking on a treadmill. First, robots have to make sense. They have to be faster and cheaper. Though they've been in the manufacturing industry for 30 years, he added, robotics are just now starting to become more accessible. They'll eventually move into business applications, like elder care in hospitals or nursing homes, or restaurant service. And then they'll hit the home. Even then, it may be just pieces of a humanoid — the arms to help wash dishes, the legs to carry heavy objects.
According to the International Federation of Robotics, in 2013, robot sales increased by 12% to a record 178,132 units. Most of those were sales of industrial robots to the automotive, chemical, rubber, and plastics industries. China (who did not end up representing at DRC) had the biggest supply of robots, with 20% of the market share, but Japan, Korea, Germany, and the US — all competitors in the DRC — had high numbers, too.
And the point of DARPA, as Pratt pointed out, isn't to bring things to market. It's to show what's possible.
On Saturday night, the winners were announced: The first place winner was Hubo, developed by Team Kaist of Korea; second place went to Running Man, developed by Team IHMC Robotics of Pensacola, Florida, and third place went to CHIMP, out of Carnegie Mellon University.
Of course, everyone wanted to win the big money for their robotics programs. HKU probably wanted to be able to afford another robot hand. MIT probably wanted to make their cheetah run faster. But the overwhelming echo during the weekend — before and after the checks were handed out — was that they all have the same goal: to further this field, keep it open source, and advance this incredible research.
As Dan Lee, a team lead from UPenn, put it — the biggest competition isn't each other. It's science fiction.
This story was updated on June 8, 2015.