ARTICLES :: "The New Mobile Infantry" (p4) :: Wired


	ARTICLES By Publication :: By Date

The New Mobile Infantry
(continued 4/4)

The PackBot, the HMTM, and all of the SAIC devices were funded by Darpa and guided by what those in the industry know as Blitch's Five Imperatives: A TMR must be able to get back on its feet when it has fallen. It must be able to recover from communication loss. It must know where it is. It must be tamperproof. And it must be able to maneuver around complex obstacles.

According to Blitch, no single tactical robot meets all five imperatives yet. But he has seen a steady evolution. "First you had radio control," he says, "where there was a full view of the vehicle at all times, and you dictated its every move." Next came tele-assisted bots, which are still guided by a human but can venture out of sight because they employ video, audio, and other sensory feedback. Tele-operated units can maneuver independently, asking questions only when they are confused. The final step, says Blitch, is complete autonomy, meaning that the robot will carry out a mission according to a set of predefined parameters, without step-by-step human guidance.

Watch Raptor move around the SAIC parking lot - controlled by Jim Hamilton and a laptop, and chased by an engineer ready to yank out the power cord if things go wrong - and you know that complete autonomy is a few years away. Currently, a typical TMR requires three or four handlers. Ultimately, researchers want to invert the ratio, allowing one soldier to control multiple autonomous robots. "Then you've achieved force multiplication," says Ron Arkin, a professor of artificial intelligence, computer vision, and mobile robotics at Georgia Tech who's written software for the Darpa program. "You could have 10 people on the battlefield doing what once took 40 soldiers."

The biggest challenge between, say, the PackBot and complete autonomy is software. It's easy enough to add another sensor; it's much harder for the robot to know how to interpret the data that sensor collects and how to integrate it with other incoming data. A tactile sensor, for instance, can "feel" uneven terrain using a series of predetermined algorithms. When the TMR discovers it's driving over rough ground, it reduces speed. Now, imagine that while the bot's sensors are detecting and responding to the surface, other sensors realize that the robotis being shot at. Should it continue slowly or speed away?

Hamilton is developing a software suite that can accept input from dozens of sensors and decide on the best course of action. Analogous to the human brain, ATAC (autonomous terrain adaptive classifier) employs an arbitrator that examines the incoming data - which might indicate gunfire, darkness, water, bioweapons, or irregular topography - and decides whether to stay and fight or run for the hills. When Hamilton sketches ATAC's decisionmaking process onto a whiteboard in the SAIC conference room, the diagram resembles the schematics of a Ponzi pyramid. At the base, there's an array of sensors, each uploading data to the next level. A more sophisticated bank of sensors at the pyramid's midsection sifts through the raw data and generates a finite set of directives. Finally, at the top, ATAC evaluates the refined readings and makes an "educated" judgment.

There are plenty of critics who doubt that software solutions like ATAC will match the decisionmaking power of the human brain anytime soon. "Autonomous robotic weapons won't demonstrate human intelligence until machines pass the Turing test," says Ray Kurzweil, author of The Age of Spiritual Machines. Other naysayers point to more basic hardware problems that must be solved. Most TMRs are smaller than tanks and less agile than humans, which can turn a molehill into a mountain. "The current wheel and track technologies are going to run into limitations," says Prasanna Mulgaonkar, the director of SRI's Advanced Automation Technology Center. He proposes a biomimetic solution: They could hop or slither or fly. Power is another stumbling block, which Georgia Tech's Arkin hopes fuel cells might solve.

"I still have not seen anything that can go where I had to go in Vietnam," says Bill McBride, a retired Marine lieutenant colonel who ran reconnaissance missions along the DMZ. McBride is a principal engineer at the Southwest Research Institute in San Antonio, Texas, where he runs the country's only independent test facility for tactical robots

Evaluators race the robots in circles on a 6,000-foot paved track until the bots either break or run out of juice. "Only when they survive the on-road track do we take them on the harder stuff," says McBride, whose offroad course includes a variety of mud bogs, water obstacles, steep culverts, rock beds, and a series of movable ramps covered with sand, gravel, and loose pipes. With each successive generation of tactical robots, McBride adds new challenges to the course. He's developing an experiment that will test how well the robots can avoid detection by a pack of aggressive dogs.

"In order to fit a TMR in your pack, you have to take out something - like ammunition - so the trade-off better be worth it," says McBride, who seems skeptical that the machines will ever be good enough. Nevertheless, he admits that a PackBot might have been handy in the battle for Hue City, a bloody house-to-house fight that cost 142 American lives and left 847 wounded. "We took a lot of casualties just trying to cross the street," says McBride.

That thought is echoed by retired colonel Mac Dorsey, who's now a program manager at Systems Planning Corporation, which provides technical and logistics support to Darpa's TMR program. As Dorsey says, "Using robots to search Vietcong tunnels would have been a much better solution than the state-of-the-art technology at the time - a very gutsy soldier."

While useful for discrete tasks, robots will never replace the well-trained soldier. "I think of them in the way a soldier thinks of his rifle: If it helps him or her get the job done, great. If not, leave it at home and take something else," says Scott Fish, the program manager of Darpa's Tactical Technology Office.

Blitch agrees that robots will never replace humans. "We are risk junkies," he says, arguing that soldiers want to be thrown into danger, not kept out of it. Still, looking at the TMRs in development today and listening to the scenarios of Blitch and other researchers, it's easy to imagine battlefields where soldiers are rarely placed in harm's way. And that makes you wonder whether we will be more inclined to go to war knowing there's less chance of losing human lives. When I ask Blitch if the robots will encourage violent solutions to political conflicts because - like the cruise missile or B-2 stealth bomber - they'll disengage us from killing, he falls uncharacteristically silent.

"A robot is not a weapon," he says, after a moment or two. "It can save someone from a sniper's bullet or be used to clear land mines all over the world." That's not to say that he doesn't wake up at night with visions of Terminator 2 replaying in his mind. "Creating machines to fight wars might indeed create more war ... even robot wars," he says. "And I don't want to go down in history as the father of weaponized robots."

In fact, he may go down in history as the first soldier to put tactical mobile robots to the test. In mid-January, four months after his unauthorized, post-retirement mission at the World Trade Center, Blitch was called back into active duty - with orders to assemble a team of robots for the mission.