The new, U.S. Army–funded forklift robot recognizes voice commands, can learn the layout of a makeshift warehouse and operate out of doors. It may be able to save lives on the field and make warehouse work more efficient at home.

Robots are becoming more common—they may not have scary lasers and godlike artificial intelligence, but they're taking over all the same. The proliferation of autonomous machines—the robot uprising, if you will— has brought nothing like the hostile army feared by science fictions writers. Rather, uncomplaining machines have taken over the jobs we'd rather not do: disarming bombs, barging into hostage standoffs, and cleaning floors and gutters. Now, MIT's latest robot tackles another unsavory task: driving forklifts in a war zone.

Using onboard laser rangefinders and navigation algorithms adapted from MIT's entry in DARPA's Urban Challenge robot car race, the semi-autonomous forklift is able to familiarize itself with a particular depot or warehouse, and then follow orders given by a remote human supervisor. Footage from the robot's forward-looking video camera is displayed on a tablet PC, which serves as a wireless controller. The operator can issue commands with a stylus, tapping on a particular pallet where the forks should be inserted, but the robot also responds to voice commands. The project is based at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL), and is funded by the U.S. Army Logistics Innovation Agency, which explains the proposed military application. According to CSAIL, forklift operators in Iraq can be compelled to leave their vehicle and head for cover up to four times a day.

By sticking a human operator in a bunker or other remote reinforced building, the business of unloading and loading supplies from trucks can continue safely and without interruption. After all, logistics in general, and supply chains in particular, are the backbone of any war.

But the potential for this robot help with warehouse work here at home isn't hard to fathom. Segway showed off an omnidirectional robot designed to operate in tight spaces, such as cramped warehouses, at last year's Robobusiness conference in Pittsburgh. Using "mecanum wheels," which are essentially wheels comprising tons of tiny rollers, the RMP 400 Omni can turn in place and glide across smooth floors while supporting more than 400 pounds on its flat back. So far, the Omni—both the 400 and scaled-down 50—has been picked up by researchers, including one team at the Georgia Institute of Technology, but hasn't entered the workforce.

Massachusetts-based Kiva, however, has already sold hundreds of its Ottoman-size shelf-moving robots to companies like Staples and Zappos. By automating the process of finding and ferrying specific items (or rather, carrying the entire shelf over to a human attendant) in a specially designed warehouse, Kiva's bots can process more items than their human counterparts can.

At the end of the workday, though, the Kiva and Segway bots are highly specialized machines, and pretty dumb. The Omni's wheels are confounded by even the slightest curb or incline, and Kiva's bots are essentially microorganisms buzzing around an expensive, custom-built robot warehouse.

MIT's heavy lifter, on the other hand, is a modified 3-ton Toyota forklift, designed to operate outdoors and indoors, in supply depots that are hammered together in a few hours. There's no magnetic tape lining the floors, no hive-mind processor holding its hand. "You put the robot in, and the robot just has to deal," says Seth Teller, a professor of computer science and engineering, and project lead.

Upon arriving at a depot or warehouse, the forklift would be given a guided tour of its environment, with the operator telling it things like, 'This is pallet two, this is pallet four, this is the commandant's lawn, don't drive there.' The robot, using voice recognition and an array of learning algorithms, pieces together its new world, tying words to places, places to actions, with everything positioned spatially. If CSAIL can pull it off, this forklift will be smarter than your average forklift, and smarter—in a real, relatable way—than most robots. "This approach makes things really hard, but if you can get it to work, then instead of just working in one of the 10 instrumented warehouses in the world, it could work anywhere," Teller says.

Nearly all of the forklift's actions are carried out autonomously, but if it can't find a particular pallet, or has trouble squeezing through a given spot, an operator can jump in and take over. There are no switches or buttons to hit—seat-occupancy sensors kick in, and the machine relinquishes control. It's a minor detail, but that sort of seamless backup might be the key to functional AI. Imagine a robotic car that drives autonomously in traffic, but transfers control when it's time to park, or at accelerating above 50 mph. Or even an upright vacuum that goes about its chores, but doesn't resist when you grab the handle to guide it through a cluttered bedroom.

The robot's voice-recognition and learning algorithms have the potential to revolutionize the way robots think, and the way we interact with them. Teller calls it "embodied voice recognition," where the machine attaches words to its environment, unlike the database-searching, disembodied recognition found in automated customer service. We call it the best kind of robot—smart, but not too smart.

Adding drive-by-wire capability and autonomy roughly tripled the cost of a $25,000 3-ton Toyota forklift, but as Teller puts it, "If it saves a couple of soldiers, it's worth a hell of a lot more than triple." He hopes eventually to roboticize larger forklifts, which can cost as much as $500,000, for less than double their unmodified price. And while there's no timeline for deploying robotic forklifts to Army depots, Teller plans to have a functional prototype ready by the spring. The short-term goal is to get the forklift working, and into dangerous environments.

Researchers at the Computer Science and Artificial Intelligence Laboratory have been testing a system designed to load and unload trucks autonomously, to be used in dangerous places such as war zones. (Photograph by Jason Dorfman)