UNDERGRADUATE ADMISSION

John Mangels
Plain Dealer Science Writer

Hunting Valley -- Dexter is learning to drive.

The setting is a farm road, a nice, safe place for a beginner. Just snowdrifts and empty fields -- plenty of room in case a first-timer hits the gas instead of the brake.

Dexter doesn't need the buffer zone. With a throaty buzz, the auto smoothly accelerates. It tracks an arrow-straight course, then crisply stops at an appointed spot 44 yards down the path. Dexter makes the run twice before getting a little disoriented and ending the session.

Not bad for a robot.

There's no human driver in this strange- looking car, not even a place for one -- no seat, no steering wheel. In fact, the driver is the car.

Dexter (a nod to "dexterous") is an autonomous vehicle. Bristling with sensors, crammed full of computers, it's designed to operate completely on its own, with the goal of driving at least as well as a person would. Dexter will make its public debut this weekend at the Cleveland Auto Show.

Later this year, Dexter's creators -- a brash, overachieving young team of more than 50 engineering and computer-science students and professors from Case Western Reserve University -- aim to win an international contest. To do so, their car must navigate a 60-mile mock urban course filled with unfamiliar roads, oncoming traffic and unexpected obstacles.

If they succeed, the members of Team Case will snare a $2 million prize and respect in the highly competitive world of automotive robotics. They also may have a hand in designing future military transports whose drivers would lose only circuit boards, not limbs or lives, if hit by a roadside bomb. Eventually, the technology should make its way into commercial vehicles.

"This is the wave of the future," said team co-leader Roger Quinn, a Case mechanical and aerospace engineering professor. "Robotic vehicles are going to happen, in the military first, and then on our roads."

Team Case faces stiff competition. Among the 89 entrants in the Department of Defense-sponsored challenge are robotics powerhouses such as MIT, Stanford and Carnegie Mellon University. Some academic contestants have partnered with big automakers and computer firms; one top contender, Team Oshkosh, itself manufactures heavy-duty military trucks.

As if things weren't tough enough for the Clevelanders, 11 of their competitors qualified for up to $1 million in development money from the Defense Advanced Research Projects Agency, or DARPA, the military think tank that dreamed up the challenge.

But Team Case has a couple of major advantages. One is the university's 20 years of experience building "biologically inspired robots" - electronic creatures that mimic the capabilities and survival strategies of insects and animals. Case labs are crawling with mechanized ants, moths, crickets, cockroaches, even a trotting robo-dog.

The other big plus is Dexter himself. (Team members refer to their vehicle as "he" rather than "it.") Conceived by ENSCO Inc., a Virginia engineering firm, the first-generation Dexter competed in the military's last robotic-car challenge, a brutal 131-mile sprint through the Mojave Desert in 2005. Felled by a flat tire, Dexter finished sixth.

Case's engineers had considered entering that 2005 challenge, but it didn't seem challenging enough, robotics-wise. The vehicles basically had to be only smart enough to track a string of GPS coordinates (programmers liken it to following bread crumbs). Precise steering and mechanical parts strong enough to survive the punishing desert course were more important than brains.

But the 2007 Urban Challenge was a different story. Congress has mandated that one-third of the military's combat fleet be unmanned by 2015. DARPA, the catalyst for innovations such as stealth aircraft and the Internet, needed to goose the pace of progress for autonomous vehicles.

"It was an area that could use a jump start," said DARPA program manager Norm Whitaker.

An urban course, with four-way intersections, traffic, pedestrians and surprise hazards, would be a lot closer to what a robotic Humvee might encounter in, say, downtown Baghdad. It also would require a substantial leap in artificial intelligence, the science of creating machines that can approximate thoughtful, purposeful behavior.

ENSCO and Case representatives met at a networking event for the Urban Challenge last May. ENSCO had Dexter but was uncertain about entering the race again, knowing that the vehicle would need decision-making abilities it lacked.

"We said, 'We don't have a car, but we do biologically inspired robotic things,' " recalled Team Case co-leader Wyatt Newman, an electrical engineering and computer science professor. "It was just a very good match of capabilities."

ENSCO lent its $500,000 car and sent along six of its team members to advise Team Case while also earning advanced engineering degrees here. Case Provost John Anderson enthusiastically backed the Dexter project, recognizing it as something the university and the community could rally around, not to mention its potential for prestige and technology spinoffs. The university has contributed about $240,000; private donors have chipped in $300,000.

Newman and Quinn recruited students for the team, though they didn't have to work too hard. The project not only provides class credit and fodder for research papers, it's a budding engineer's catnip.

"It's a nerd magnet," laughed Newman, who counts himself as one.

"That's the easy thing, getting [students] excited," added Quinn. "These are people that love robots; they dream about robots. Often that's what got them into engineering."

Brad Hughes is a freshman with a heavy class load, but he spends 25 hours a week tinkering with Dexter's sensors and mechanical systems. "It's a great opportunity for research," the 18-year-old said. "I've learned a lot, maybe more than in some of my classes."

Team Case's biggest task is enhancing Dexter's brain - stripping out the existing software and writing thousands of lines of new computer code to enable it to make the right decisions and complete the urban course.

Deciding how to program Dexter, the team confronted a fundamental schism in the artificial-intelligence community. It involves differing views of what intelligence is and how to try to re-create it in machines.

The classic AI approach, with its roots in the earliest computer chess-playing programs written in the 1950s, attempts to assemble sets of logical rules that define any possible condition. Faced with a decision, the computer sifts through all the rules until it finds one that applies to its current circumstances.

That's fine for something like chess, where there are a large but not infinite number of potential moves. In fact, Deep Blue, a chess-playing computer designed by IBM, was able to ponder 200 million possible board positions per second, allowing it to "think" up to 30 moves ahead. Deep Blue beat world chess champion Garry Kasparov in 1997.

But real life, with all its variables and uncertainties, doesn't lend itself well to classic AI. How could a rule-writing programmer possibly anticipate everything? Driving, for example, has rules of the road, but they don't fit every situation. No two scenarios are exactly alike. And what if you encounter another driver who isn't following the rules?

New-wave AI accepts that rules can't cover everything. Its marching orders are more general: "Do the right thing." A cockroach is a good example of this approach. Its tiny brain can't comprehend what humans are, yet it knows to scurry away if we flip on a light or step too close. It doesn't understand reproduction yet can find a mate by tracking its odor.

"It's making sensible, intelligent decisions and acting on them," Newman said. "And it's doing it rapidly. Not optimally, but optimality isn't the point. Doing what's necessary to survive is important."

The roach may encounter unfamiliar situations, but its ability to respond based on some general guidelines is good enough to get by.

Team Case is adopting that behavioral approach with Dexter. Its student programmers have defined about 20 key driving "behaviors" - stopping, accelerating, staying in a lane, following another car at a set distance. "Moods" string together these behaviors to do something useful, such as passing.

Above moods in the programming hierarchy are "mood selectors," which incorporate an awareness of Dexter's overall start-to-finish mission. They function like emotions. "A mood selector says, 'I know I want to turn left. Are conditions optimal? How do I prepare?' " explained graduate student Andrew Horchler, 29, who leads Dexter's behavioral control team.

Each intended action is checked against Dexter's overall goals of driving safely, legally and efficiently. Cameras, radar and other sensors constantly monitor surrounding conditions and update Dexter's position relative to other objects.

Any parent or driving instructor knows that teaching a novice to operate a car is a time-consuming, trial-and-error job. In a windowless eighth-floor classroom adorned with empty pop cans, circuit boards and a plant in need of watering, Team Case members are working nearly 24/7 to write code and fine-tune Dexter's operating system.

Before trying out driving instructions on the actual vehicle, students check them on a video game-style simulator. Eventually, when Dexter's brain gets smart enough, it will relay proposed actions to a human driving the team's van on a test course.

The engineers will scrutinize any differences between Dexter's commands and the driver's decisions, since those could indicate programming flaws.

Already, Dexter has successfully negotiated a route of one-way paths on Case's Hunting Valley farm, picking the appropriate directions, following the lanes and stopping at the right places.

The first major Urban Challenge hurdle is April 13, when teams must submit a video showing their vehicle driving itself around an oval track and safely steering around an obstacle car.

DARPA officials will visit qualifying teams this summer to personally observe a more-complex driving test, which includes moving traffic. Those who pass go to the finals this fall. The location is secret, to prevent teams from scouting the site.

If Team Case members are apprehensive about what's ahead, they don't show it. Their weekly meetings are a mix of good-natured jokes and slick PowerPoint presentations charting their progress. They check out their competitors' Web sites from time to time but stay focused on their own tasks. Being an underdog has its advantages.

"The highly funded, highly seeded teams are probably a lot more concerned and nervous than our students," said electrical engineering and computer science professor Michael Branicky. "For us, it's about building capabilities and building a robot that could possibly win. For them, they also have the larger task of not failing."

© 2007 The Plain Dealer