The Surreal Agony of Waiting for Robots to Do Something
There is something about the DARPA Robotics Challenge Finals that felt like watching the Masters.
Engineers with Japan's Team NEDO-JSK ready their bipedal humanoid rescue robot, Jaxon, at the recent DARPA Robotics Challenge Finals. Photo: Justin Staple/MOTHERBOARD
Jaxon tried opening the door. But nothing happened.
"Do something!" someone yelled. Do something. It became the phrase of choice at the recent DARPA Robotics Challenge Finals, where Jaxon, a bipedal humanoid rescue robot, competed against robots helmed by teams from around the world. What usually happened next, after someone would yell at a robot to do something, is the robot either did nothing, did something very, very slowly, or fell.
Then it happened: Jaxon fell, and couldn't get up. A handful of engineers who service Jaxon scrambled to place their hulking, fallen machine on a stretcher. A crowd of several thousand people, gathered over the weekend in a sprawling fairground in Southern California, laughed nervously while the engineers hauled Jaxon off for repairs.
There is something about the DARPA Robotics Challenge Finals that felt almost surreal. The teams responsible for creating the top ground-based, semi-autonomous rescue robots in the world, from US universities to labs in Italy, Germany, and Japan, had convened at the FairPlex, home of the Los Angeles County Fair, to demonstrate how their mobile robots might perform in humanitarian aid and disaster response situations that are otherwise inhospitable to humans. Why send humans into dull, dirty, or dangerous situations like nuclear reactor meltdowns, infectious disease hot zones, and burning buildings, advocates for these technologies argue, when we can send robots? This is the DRC mantra, one that makes no mention of one day weaponizing any of the technologies developed throughout the challenge.
The Finals were the culmination of a multi-year program conceived by the Defense Advanced Research Projects Agency, the Pentagon's blue-sky research wing, in 2012. The marathon two-day event drew nearly two dozen teams of engineers, roboticists, scientists, military personnel, and an eager audience from around the world.
The rules were such that all 23 teams competing had to operate their robots remotely and wirelessly as they attempted to navigate them through obstacles courses meant to mimic disaster scenarios. They couldn't use tethers to stop their robots from falling, and, to make things even more challenging for the teams, those wireless connections were intentionally spotty. As a way to simulate crisis conditions like the Fukushima disaster—the catalyst for the challenge—DARPA purposefully degraded the communication links between the robots and their human operators. Think of it like talking with someone over the phone, but only receiving a miniscule percentage of the information being communicated to you.
Watch more from Motherboard: The robots are here.
Those are considerable constraints, but, despite them, I saw robot after robot walking over uneven terrain, walking up staircases, driving cars, and handling power tools with total control. Teams had one hour to guide their robots, two- and four-legged machines that go by names like WALKMAN, FLORIAN, and ROBOSIMIAN, through an obstacle course and complete a range of tasks. (There were four identical tracks, running parallel, and at any given time there were two teams running their bots on them. Each team did multiple passes throughout the two days.)
The tasks included having the robot drive a vehicle down a dirt track to the course; exit the vehicle; open a door; cut a circle (with a small saw) out of a plaster wall, and then push it through; rotate a circular valve 360 degrees; pull down a lever; navigate through a small rubble bed; and, finally, walk up a small staircase.
"It took a helluva fall"
Teams were awarded points for completing each task, and were allowed to bypass some of the tasks. Many of the teams took advantage of this, including Team Valor, a Virginia Tech robotics group that I recently spent time with while filming our doc about robots. They decided to skip the car-based challenge because their bipedal robot, ESCHER, wouldn't have fit behind the wheel in the first place. So they started ESCHER at the door task, but quickly ran into other problems.
"It took a helluva fall," John Seminatore, a graduate student and Valor team member, told me as Day 1 drew to a close.
We were tucked away in the service garage, a large hangar-like space that served as makeshift lab and repair shop for all 23 robot teams. Seminatore told me the team ran up against computer issues, and had to reset their robot. Then they inched ESCHER a quarter of the way further through the course before he went down.
It was a similar fate that befell Team NEDO-JSK, who, operating Jaxon remotely from a garage a half-mile away, likewise strained to communicate with their machine. They had it grab the door handle and push it down and forward, when suddenly Jaxon, standing as tall as an average human adult and sporting a hot red and yellow protective exoskeleton, went down. Then they took him away on the stretcher.
Not that any of this was unexpected. Before the Finals got underway, Gill Pratt, the DRC's program manager, had told reporters, myself include, to "expect to see a lot of robots falling down."
Video courtesy DARPA
Few teams made it the entire way through the course without running down the hour clock, without having to reset their robot, or without their robot falling. Often, due to the element of degraded communications, it was all three. One team, however, did have a flawless run: South Korea's Team KAIST, seen above beginning at 08:21, completed the course in under 45 minutes.
There was occasional drama, most notably when a robot would begin wobbling but manage to right itself, staving off a potentially challenge-ending fall. Tracking team standings on a jumbo scoreboard was something out of the Masters, and Pratt did liken the competition to "watching a golf match." And sometimes, things really moved slowly. During one prolonged intermission, as robots from two teams drove slowly down to the course, Elton John's "Rocket Man" swelled from the FairPlex PA. There was a lot of time spent watching, waiting, hoping for a robot to do something.
"Do something!" someone would inevitably yell.
I heard it inside FairPlex pavilion, in the infield on an aging horse track, where the robot action, if it could be called that, took place. I heard it it outside the main pavilion, as well, in a sprawling parking lot expo that kind of felt like your average tech trade show, but with a heavier military ambience and vaguely dystopian signage. A large DARPA banner hanging near a highly-trafficked intersection of the expo read: TECHNOLOGY BEGETS TECHNOLOGY. Another said: THERE IS NO FINAL FRONTIER.
There was pretty much every robot-related thing imaginable at the expo, from bomb-disposal platforms to small unmanned aerial vehicles to fun, kid-oriented challenges to get youths fired up on STEM. A pair of R2D2 droids even blorped around.
Over at the vendor booths, the expanding prosthetics and wearable bionics market really dominated the scene. Notably, and unfortunately, most of the prosthetics I encountered were only fit for man hands. But at a far end of the expo I saw an exoskeleton with an extendable gravity-balancing arm and attachable grinder. I had to try it on.
The rig was a prototype designed and built by Ekso Bionics, a wearable bionics company specializing in gait training for individuals with various levels of lower body weakness. There are currently around 100 Eksos, the company's flagship weight-bearing medical exoskeleton, in the field, a rep told me. It seemed the new prototype, of which there are currently only two, might be a play to get those overseeing manual laborers to invest in exoskeletons up front, to spare bad backs and related medical costs.
The problem for me, and people of similar stature, is apparently I am just under the ideal height (I'm 5'6), so the upper parts of the rig were a little tall on me. The key with exoskeletons, according to the rep, is to align the joints of the suit with the joints of the person wearing it. Jeffrey Vickers, the mechatronics engineer at Ekso Bionics, made a height adjustment to the frame, and cinched down the shoulder and waist straps. I took a few steps forward.
The suit made me feel quite light. I was able to fully bend my knees and walk in circles, even though my hip and the suit hip weren't perfectly aligned. It didn't follow my step as tightly as it would for someone over six feet tall (the size frame this prototype was designed to fit, I'm told) though I couldn't see it taking long to get used to, either way.
I felt like I could go clear a stand of old growth pines.
Then came the gravity-balancing arm and grinder. The suit doesn't lighten a load per se, but supports the weight of specific objects strapped to it by distributing the weight evenly around the frame. Together, the arm and grinder added between 20 and 30 pounds onto my right arm, and yet I felt nothing. The arm unlocked, and I could swerve the grinder in and out at a length of about four feet.
"Think you could work like that for eight hours?" Vickers asked. "The whole idea is you won't feel this weight."
I felt like I could go clear a stand of old growth pines.
"Now you're a worker."
Back on the tracks, FLORIAN, the bipedal humanoid robot helmed by Team ViGIR (also featured in our doc Inhuman Kind), had just crashed his car into a protective barrier. "Our robot is very strong and very big," admitted Stefan Kohlbrecher, one of FLORIAN's human operators.
This happened multiple times throughout the event, in fact. Watching numerous robots drive through weighted barriers stood in stark contrast to depictions of highly-intelligent robotic technologies depicted in recent sci-fi films like Chappie and Ex Machina.
When I caught up with Brian Lattimer, a professor of engineering at Virginia Tech and Team Valor member, he said he thought all the robots competing in the Finals "have particular positives and negatives. There are certain tasks that certain kinds of robots are better for, while robots of different designs might be better for different things. It depends on the function. I think our robot is particularly good at walking on really complicated terrain, and being automatically adaptable."
I wanted to give Lattimer the opportunity to respond to some of the criticisms to his work, as seen in Inhuman Kind, so I asked him specifically about a comment from YouTube user "Howden Gray," who said: "Of course robots will be weaponized. It's like trying to stop people from putting guns on helicopters."
Lattimer had no response to that comment. He stressed how Team Valor is developing robotic applications to help people. "To help people keep their jobs by having robots do what the elderly population can't do anymore, and trying to keep them working," he said, "trying to help people on third shifts. Trying to help people in disaster response. That's where we're focusing our research."
DARPA did not return my request for comment on any future plan to weaponize, or use nefariously any technologies first developed for the DRC.
When it was all over, the winning robot team, from South Korea's Advanced Institute of Science and Technology, was awarded $2 million by the US military. (That's them in the above video.) The second and third place teams were both American, and went home with $1 million and $500,000, respectively. Team ViGIR finished in 15th place, and Team Valor came in 19th place.
"You are going to be the vanguard of this new future … that allows us to have robots out in this unstructured environment," Arati Prabhakar, the head of DARPA, said during the awards ceremony, addressing the winning teams and a dwindling public crowd in the grandstands. "This is just the beginning."
It was hard to leave with impression that the near future of advanced humanitarian robotics is one overrun by Skynet and Terminator-style killing machines. That future looks a lot more like human toddlers learning how to walk and perceive the world around them. Advanced robotics and artificial intelligence are in a larval moment, which is equally exciting and sobering. But if the spectacular robot spills at the DRC were any indication, breaking through that stage is just not going to happen anytime soon.
Until then someone, somewhere will be yelling, "Do something!"