The DARPA Cyborg Showdown Looked Like a Golf Game Between Giant Babies
Real progress in humanoid robots isn’t going to come from DARPA's comparatively puny million-dollar awards. This is now Google's game to lose.
Despite a handsome budget, NASA JSC's Valkyrie scored zero points. (DARPA)
I was looking forward to spending my weekend at home, eating chips and avocados, staring at my computer screen. The Defense Advanced Research Projects Agency, the U.S. military research wing that’s tasked with turning science fiction into reality, was going to bring together seventeen cutting-edge walking humanoid robots to compete at a speedway near Miami. The robots all looked exactly as one would hope—complex, hulking and frightening—and they were there to prove that a machine could walk into a simulated nuclear emergency and save the day.
I was excited. And I was ready to be bored out of my skull.
“It was kind of like watching a golf match,” said Michael Belfiore, who has written a book about DARPA and traveled to the Speedway to spectate. “I don’t watch golf, but I imagine that this is what it would be like.”
When I tuned into the live Youtube broadcast, a robot about the size of a ten year old boy was standing motionless on a stepladder.
“What’s interesting about this robot,” said the announcer, “is that it’s climbing the ladder backwards.” After what seemed like a very long time, the robot—which was representing Drexel University’s DRC-Hubo team—took a single backward step up the ladder. The crowd cheered. I watched as it paused again for an equally long time before taking another step. When it reached the second-to-last step, the robot slipped off the ladder. It would have smashed into the ground had it not been for the harness attached to its head.
“You could hear the crowd saying ‘AWWW,’” Dr. Paul Oh, the director of the Drexel team told me over the phone later.
And that’s pretty much as exciting as it got.
A DARPA preview video, "Robotic Expectations"
A little later, the feed switched to a confused-looking Boston Dynamics Atlas robot, which was being run by the TracLabs team, standing in front of a large valve with a big sign that read “CYANIDE.” Apart from a rotating gimbal on its head, the robot wasn’t moving. Throughout the broadcast, the announcer tried his best to keep things exciting: “As you can see, we have quite a crowd here today!” he said. “If you haven’t already, download our app!” The fact of the matter was that, like golf, this wasn’t an action sport.
When you watch a video of the Boston Dynamics Atlas robot scrambling over rocks in the company’s now famous viral videos, it seems like we’re just around the corner from a Ghost in the Shell moment. And yet, at the Robotics Challenge, it became clear that we’re still a long way off. Out of a whopping nine and a half hours of footage from one day of the "Red Terrain" challenge, things only start to get interesting around the seven hour mark. (But not that interesting.)
Nine and a half riveting hours of the "Red Terrain" competition
“It’s like when my baby took her first step,” Dr. Oh told me. “Yes, she fell down, but I was elated.”
While robots appear to possess great dexterity and poise in the perfectly controlled environment of a laboratory, as soon as you put one into the unpredictable real world, its capacity do to anything useful is greatly diminished. Nevertheless, for the world of robotics, this event was like the World Cup.
“It’s like when my baby took her first step,” Dr. Oh told me. “Yes, she fell down, but I was elated.” That being said, he did admit that the Challenge was perhaps a little too slow-paced. “The public was being very patient.”
This baby metaphor seemed to be a motif at the tournament. These robots have “hardly any brains at all," Gill Pratt, Project Manager of the Robotics Challenge, explains in a pre-competition DARPA video named “Robotic Expectations.” "The head is empty.” He explains that the robots have about the same level of competence as a one-year-old. While he is explaining this, the video switches to a clip of a child playing at one of the team garages, in case we had forgotten what a human child looks like.
Of course, the path to these baby steps is a long giant leap. Billy Howell, a multimedia specialist from the Florida Institute for Human and Machine Cognition, was delighted to watch his team's robot, a modified Boston Dynamics Atlas, walk across the terrain, however slowly (it came in second, with 20 points total).
"This is an extremely hard field to even begin to perform well in," Howell told me. "Jerry Pratt, our head robotics researcher, has been working on bipedal walking his entire career. He's got some pretty amazing ideas when it comes to bipedal robotic walking. IHMC wrote our own control algorithms for Atlas from the ground up, so it's been a seriously intense journey… learning someone else's robot, the worlds most advanced hydraulically powered one at that, and on top of that, how to control it."
There was one exception to the ponderous and clumsy robot heroics: Team SCHAFT. This nascent Japanese company was founded by three graduates of the University of Tokyo, which has one of the best robotics programs in the world but doesn’t permit military-funded research. If Team SCHAFT's robot were a basketball player, it would be LeBron James.
HRP-2, as it’s called, ended the competition with 27 out of 32 possible points, which was seven points ahead of runners-up IHMC Robotics. Its arms are ten times stronger than those of traditional humanoid robots, and its body is pumped full of blood-red cooling liquid, which undoubtedly adds to the badass effect. HRP-2 breezed through the eight tasks of the competition.
Tokyo University's HRP-2 robot decimated its rivals, slowly
Each time the robot completed another challenge successfully, Belfiore reported that the company’s CEO, Yuto Nakanishi, who looks like an energy-drink-fuelled mad scientist, would jump up and down and pump the air. Where other teams’ robots were hesitant, SCHAFT's robot was assertive. The HPR-2 was hardly Usain Bolt (that's how some refer to Boston Dynamics’ Cheetah). But compared to the other robots, that thing could move.
So did we just witness a watershed moment for humanoid robotics? While the media reaction was rife with Wright Brothers comparisons, there is simply no way that the event was anything as dramatic as the first manned airplane flight. It took the robots thirty minutes to walk across a couple of mounds of cinderblocks that a human, even a human wearing a pair of computer glasses, could probably skip across in thirty seconds.
But when DARPA held a similar Challenge for driverless cars back in 2004, not a single team completed the 150 mile route. The most successful team only managed to get five percent of the way there. The New Yorker’s Burkhard Bilger described it as “one of the more humbling events in automotive history.” Less than a year later, however, five teams successfully completed a much more complex 132-mile course. If the same rate of progress applies to the Robotics Challenge, we should see a much higher success rate at the finals in 2014. Or, as a DARPA info sheet put it, “we expect the robots will demonstrate roughly the competence of a two‐year‐old child.” That is, a 6’ 2”, 330 lb two-year-old child.
Yuto Nakanishi, CEO of IHMC Robotics, saluting the winning robot, HRP-2
Impressive as that may sound, the real progress isn’t going to happen through DARPA and the comparatively puny $1 million research funds that it will award to each of the top eight teams from the Robotics Challenge. It’s going to be thanks to Google.
In the past six months, Google has snapped up both SCHAFT and Boston Dynamics, whose Atlas robot was used by seven of the seventeen teams at the Challenge, in addition to six other robotics companies. This isn’t the first time that DARPA has handed the baton to Google. Following the driverless car competition, Google cherry-picked the most talented members of the teams to join its Google Drive project at Google X. Now, Google Cars are driving nearly-blind men around San Francisco. If the driverless car's progression from DARPA fantasy to Google reality is anything to go by, we should have Sergei Brin shaking hands with a humanoid robot by about 2021. (It's not yet clear who will have the stronger grip.)
All of it may have been hard for me to appreciate at a distance, through the telepresent technology of YouTube (thanks, Google). But the sense among those who participated in the Robotics Challenge was that something for the history books was taking place. With Google now bankrolling so many of the horses at the robot racetrack, they might be right.
There was another winner too, though: the robot in its slow but steady trot towards something resembling human achievement. I asked Dr. Oh, leader of the Drexel team, and its HUBO robot, what the highlight of the event was for him. “During the driving challenge, I was down field, waiting for the car. Then I heard the crowd cheering when the car started to accelerate. That energy was willing the robot forward,” he said. “I welled up a little bit.”
- Artificial Intelligence