Robin Deits shouts it not as a rallying cry, but as a warning to everyone nearby that the 6-foot tall, 330-pound Atlas robot hanging from a steel beam is about to wake up. A compressor starts to whine, pumping the hydraulic fluid that will enable Atlas to stand and move on its own. A bright orange warning light on the top of its head starts to blink. Once the team of MIT professors, students, and research assistants is convinced Atlas is ready to stand on its own, Steve Proulx lowers it to the floor using a rope and pulley system.
Inside this cavernous shed on the Massachusetts Institute of Technology campus, Atlas’s big task today is something most people do dozens of times daily without thinking: open a door and walk through it. The first time I watch the robot try, it nearly falls over, and Proulx has to pull the rope taut to keep the machine from hitting the floor. There are only eight of these robots in existence, and they cost more than $1 million each.
MIT’s Atlas bot and the accompanying software that enables it to pick up objects, use power tools, and drive a car, will be among 17 competitors next weekend at the DARPA Robotics Challenge Trials in Florida. (DARPA is the Pentagon’s Defense Advanced Research Projects Agency — the same folks who brought you the Internet.) It’s essentially the NFL Playoffs for the world’s most sophisticated robots — with a lot less hype and a lot more purpose. Eight teams will make it to the finals in 2014.
The challenge was designed to encourage university teams, private companies, and government agencies like NASA to radically improve the ability of robots to operate in the real world, and assist with disaster scenarios like the Deepwater Horizon oil rig explosion or the Fukushima reactor meltdown in Japan.
“At Fukushima, you had an environment that was built by people but became uninhabitable by people after the earthquake,” says Seth Teller, an MIT professor who leads a team participating in the robotics challenge. “Today’s robots could go in and look around, but what if we could have sent in a machine to avert the disaster?”
The challenge, which will eventually award $2 million to a winning team, involves several local players. The teams that chose to focus on writing software, rather than designing their own robots, have been using Atlas, made by Boston Dynamics of Waltham, which was purchased by Google on Friday. That includes MIT’s team and one from Worcester Polytechnic Institute that also made it into the playoffs. (WPI’s bot has its own Twitter account, and a bio that describes it as “tall, dark, and humanoid.”)
The teams have also had access to a robotic hand, dubbed Dexter, made by iRobot Corp. of Bedford. IRobot and Boston Dynamics will also be demonstrating a few existing products in Florida, like the FirstLook, a surveillance bot that can be tossed over a wall or into a building, and WildCat, a four-legged bot that can run at about 15 miles per hour.
But the real action next weekend will be the challenge itself, when robots go head to head. They will attempt to perform tasks like climbing a ladder, driving a small all-terrain vehicle, using a cordless drill, and walking through doors. Robots earn points by doing tasks successfully within a specified time frame, with bonus points awarded when humans don’t need to intervene by, say, getting a robot’s hand out of a jam or using the rope to keep it from falling over. The challenge itself was created by Gill Pratt, a DARPA program manager and former Olin College engineering professor who lives in Lexington.
Watching the MIT team conduct some final test runs last week, it quickly became apparent just how difficult it is for robots to escape from factories and move around in environments built for humans. On a computer screen in front of Deits, I could see one of the ways Atlas perceives the world, using its on-board laser scanner: as a colorful “point cloud” that looked like a digital version of a Georges Seurat painting. Deits and Pat Marion, a fellow operator, had to literally lay out the footsteps Atlas would follow to approach the door, and point out where the handle was so that it could properly grasp it. With the door opened, the robot had to turn 90 degrees and walk sideways through it; the doorway was too narrow for Atlas’s broad aluminum shoulders. After the first failed attempt, Atlas made it through successfully — after about 10 minutes of painstakingly slow maneuvering.
I asked Teller and Russ Tedrake, another MIT professor, about the task they found the toughest. They were both quiet for a while, wondering if it was strategically unwise to share that information publicly. But then they agreed that driving the ATV, a Polaris Ranger, and especially getting out of it, was close to the top of their list.
Whichever teams emerge victorious from next weekend’s event in Florida, one thing is certain: DARPA is accelerating robotic evolution, both in academia and the private sector.
“DARPA’s intent was to try to push the state of robotics technology, to the benefit of disaster response,” says Chris Jones, director of research advancement at iRobot. “They’re trying to tackle some really difficult problems, things that are 10 or 20 years out.” For iRobot, DARPA research funding enabled the company to develop a nimble three-fingered robotic hand that can pick up objects ranging from a basketball to a pair of tweezers to a CharlieCard lying flat on a table. In a business environment, Jones says, the hand could help one of iRobot’s Ava videoconferencing robots open doors, or push elevator buttons to make its way around an office building.
“The obvious analogy goes back to the 1980s and personal computers, when individuals first got access to computing technology,” says Teller. “We’re not quite there yet with robots, but we will see more of these machines in homes, hospitals, public spaces, and workplaces.” The DARPA Robotics Challenge, as Teller sees it, “is a precursor to introducing these machines into the home. After all, my kids’ rooms usually look like disaster zones.”