For all the attention they get, humanoid robots tend to be a pretty shallow bunch. Honda's Asimo dances, shakes hands, and occasionally serves tea. Toyota's series of Partner Robots can play musical instruments and guide visitors around one of the carmaker's facilities in Japan. A range of less famous models in labs around the world grab headlines by gripping objects without destroying them, or walking a few steps without tipping onto their extremely expensive heads. Humanoid bots are the celebrities of the robot world.

Which is why the unveiling this morning of Robonaut2—or R2—a collaboration between General Motor and NASA's Johnson Space Center, is such a milestone. R2 is the direct descendant of Robonaut, a humanoid model designed by NASA to assist astronauts during spacewalks (or extravehicular activities; EVAs, as the agency calls them), planetary exploration or any mission that could use an extra pair of dextrous hands. NASA intentionally avoided the complex, expensive business of two-legged mobility, instead fitting the robot with a single leg, designed to fit into the foot-restraints used by astronauts during EVAs. The robot could also be mounted, Centaur-like, on a wheeled platform. Robonaut never made it into space, but starting in 2007, General Motors embedded a team of their engineers with the existing Robonaut team at Johnson Space Center in Houston, to help design the robot's successor. GM also provided funding for the project, a move that, given NASA's current budgetary reshuffling, could be visionary in hindsight.

So why, exactly, would an embattled automaker devote its dwindling resources to a robot designed to clamber around spacecraft or motor across other planets? For the same reason GM has always been interested in robots: to build cars. "We had a common agenda with NASA," says Allen Taub, vice president of global research and development at GM. "They wanted to make a robot that could work next to an astronaut," he says. "The question we wanted to answer was, 'How do I make a robot so it can work with operators, without all of the safety precautions and cages?'" As they go through their automated routines, industrial assembly bots are inherently dangerous to be around. And according to Taub, installing cages and other safety measures often costs more than the robot itself. "This robot can be going through its paces, and if you just hold your hand up, it hits your hand and stops," he says.

On paper, that's not a new feature—NASA designed the original Robonaut to be controlled directly by a human, through teleoperation, or to follow various human-given commands, which the engineers called supervised autonomy. R2's engineers have focused more on that autonomy, but according to Robonaut2 project manager Myron Diftler, the advantage of working with GM was the company's experience with robots that can be used continuously, and by people without advanced degrees. "Their focus on durability and ease of use is different from NASA's," Diftler says. NASA robots are typically operated by highly-trained personnel, who issue various complex commands. And as robust as the Martian rovers appear to be, space-bound bots are designed to perform a given action hundreds of times. GM's experience with industrial robots helped the R2 team simplify the control interface—meaning less time and money spent on training, and less babysitting the bot by crew members in space or on Earth.

GM's goal in co-developing R2 is to eventually install similar systems in its plants, performing the kind of repetitive, ergonomically difficult jobs that might injure a human operator. Vision sensors in the robot's head, as well as pressure sensors in its fingers, allow it to manipulate parts with near-human precision. The biggest upgrades from the original Robonaut are R2's thumb, which now have four degrees of freedom (as opposed to three), and its overall speed, which have improved by a factor of four. One result of all of this engineering is the kind of breakthrough only a roboticist would swoon over: R2 can use both hands to work with a piece of flexible material. If that sounds simple, consider the amount of sensory data, cognitive processing and physical dexterity needed to manipulate something that flows and bends in your fingers. In the series of baby steps that comprises robotics, R2 is leaping.

Still, the two existing R2 prototypes are still essentially legless—GM has no need for a bipedal robot awkwardly swaying through its plants, and NASA plans to fit the robot with at least as many mobility platforms as its predecessor. R2's lower half is intended to be modular, and so is its redesigned head, which could fit a variety of sensor suites, depending on the mission or environment. Of course, until the agency's budget is sorted out, Diftler can't confirm what those missions will be, or when the robot could be deployed. Which means the robot, or some version of it, is more likely to show up in a GM plant before leaving the planet. Taub considers Robonaut2 a concept demonstration, analogous to what the company proved when its robotic Chevy Tahoe, Boss, won the DARPA Urban Challenge in 2007. As GM continues to work with NASA, Taub hopes to have a humanoid robot on the line at a pilot plant within 3 to 5 years (R2-related improvements, such as in vision processing, could make it to existing plants even faster). In fact, he sees a similar overall timeframe between autonomous driving and autonomous humanoid workers. The comparison is easy to make—GM provided the first glimpse of the robot chaffeur with Boss. Now, Robonaut2 could be the first blue-collar robot. Taub and Diftler also envision R2 being adapted for workplace environments where bulky protective gear limits human dexterity, such as clean rooms or nuclear facilities. Wherever it ends up, R2 represents a shot across the bow for the mechanized humanoid dilettantes at Honda and Toyota."The way we refer to it in the company, is that we've designed these robots to do work," Taub says. "Although it is fun and exciting, this is not show and tell."