Self-Navigating Vehicle

It's shortly before dawn, and the handful of early-morning commuters on the fog-shrouded suburban highway don't see the deer meandering across the road. Luckily, though, their cars see the animal. In an instant, the closest vehicle quickly applies its brakes and turns its wheels, steering around an otherwise imminent collision. It then sends warning messages to oncoming traffic, as well as to the vehicles behind it, which dutifully apply their brakes and slow to a near-crawl as the deer passes.

None of the drivers, however, is disturbed by the near miss. A few are too engaged in their morning newspapers; a couple more are snoozing for a few more minutes before arriving at the office. All are blissfully unaware of the incident because they, the "drivers," aren't driving; they're being chauffeured by their self-navigating vehicles.

Sound impossible? Many automotive experts don't think so. The technological pieces needed for a self-navigating vehicle are already falling into place, they say. But it will take at least two to three decades before those pieces will be assembled into a car that drives itself.

"We will get there, step by step, not because someone is developing a self-navigating car, but because the enabling technologies are starting to show up for other reasons," noted Dr. Peter Schulmeyer, director of strategy and marketing for Freescale Semiconductor Inc. (Austin, Texas). "One day, someone will realize that we are almost there, and they'll take that last step."

"A lot of people like driving," Schulmeyer said.

Still, there are powerful reasons for engineers to pursue autonomous vehicle technologies. Millions of commuters, for example, now spend two unproductive hours per day, and more than 500 unproductive hours per year, driving back and forth to work. Moreover, many of those drivers die in crashes.

Experts say that most of the 40,000-plus annual U.S. highway fatalities are caused by some form of inattention, ranging from an inopportune turn of the head to a driver simply falling asleep at the wheel.

Engineers aren't trying to solve the entire problem; rather they are developing the technologies piece by piece. The first piece, a so-called "adaptive cruise control" system that uses forward-looking sensors to monitor objects ahead of it, is already being put into production vehicles.

Similarly, "lane keeping" systems, which alert drivers when they are wandering out of their lanes, are also reaching the market now. The third piece of the puzzle-a "collision avoidance" system that can steer a vehicle out of an imminent collision-is still under development.

In Darpa's Grand Challenge, self-navigating vehicles drove over sand dunes and through rivers.

To reach complete autonomy, however, requires going far beyond those three pieces. Vehicles will need GPS-based navigation systems, combined with accelerometers and gyros for dead-reckoning navigation, to help map their routes.

In addition, they will need electronically controlled steer-by-wire, brake-by-wire, throttle-by-wire and suspension-by-wire systems to better enable autonomous steering, braking and accelerating. They will need faster and more powerful computers-a multitude of them-to deal with the extraordinary amount of data being processed within the vehicle. And they will need, of course, better sensors-CMOS cameras and radar-to "see" the world around them.

"The most important technology will be the sensors, which will give us the capability to see and understand what is happening around the vehicle environment," Schulmeyer said. "The radar systems today are enabling adaptive cruise control, but with these sensors, it will become possible to create features, such as collision avoidance. If the car can slow down as part of adaptive cruise control today, then maybe in the future it can apply its brakes even harder, or take control of the wheel and become a safety system."

In March, the agency made its most serious effort to date at developing the technology by sponsoring a $1 million, winner-take-all driverless 142-mile race across the desert in California and Nevada.

Participants in the race said they believe that engineers are much closer to developing autonomous vehicles than the public realizes.

"It's inevitable," noted Bruce Hall, president of Velodyne Acoustics Inc. (Morgan Hill, Calif.). "There are a lot of obstacles between now and the time you'll be able to program your vehicle to drive you to work, but we're definitely going to get there."

Vehicles could improve safety by communicating wirelessly with one another when they are stopping or reaching an intersection.

Hall and his brother, engineer Dave Hall, came in third place in Darpa's Grand Challenge, going approximately six miles before their vehicle got hung up on a small rock. The Halls' team, known as Team DAD (Digital Auto Drive), combined radar sensors and CMOS cameras with two Texas Instruments 6400 DSP chips to create a stereo vision system. The digital signal processors, operating at 1.1 GHz each, built a terrain map for the vehicle to follow.

"We were chunking a 600- by 400-pixel map at a rate of 60 times per second, and building a terrain map, so the vehicle would know where the objects were, and where the road was," Hall recalled.

In addition to the sensors, Team DAD and other competitors incorporated on-board inertial measurement units (IMUs), which used accelerometers and gyros. The IMUs measured changes in each vehicle's heading angle, then compared those values to GPS navigation data to determine whether the cars were on course.

Processing data from all those systems, particularly while the vehicles were traversing sand dunes and crossing rivers, required extraordinary amounts of computing ability. Referring to their vehicles as "supercomputers on wheels," many of the entrants packed as much as 60 pounds worth of on-board computers.

A vehicle from Carnegie-Mellon University (Pittsburgh), for example, incorporated a four-way Intel Itanium-based parallel processing system and three dual-Xeon processor setups. Similarly, a team from SciAutonics (Thousand Oaks, Calif.) employed a PowerPC-based vehicle control computer, three ruggedized laptops and a separate industrial computer, all connected over an Ethernet data bus, to handle inputs from navigation systems, ultrasound sensors and infrared lasers around the vehicle.

Signals from radar sensors and cameras serve as the eyes of a self-navigating vehicle.

Because much of the race took place off-road, the vehicles had to be capable of reading the terrain, navigating around rocks and trees, and recognizing steep drop-offs.

"One possible scenario for these vehicles is that they will drive right off a cliff," noted Charles Reinholtz, Alumni Distinguished Professor of Mechanical Engineering at Virginia Tech University. "Perceiving a drop-off is very difficult for a computer, because, unlike a tree or rock, it's a 'negative obstacle.' "

Reinholtz added that Virginia Tech's team of 30 students and nine professors named their vehicle "Cliff" with such possibilities in mind.

The challenges of "terrain comprehension," however, are not limited to off-road races. According to engineers, better software will be needed in order to understand the obstacles that drivers face every day.

"The ability of humans to drive down the street and perceive the head of a boy behind a parked car is very impressive," Hall said. "It's going to be very, very difficult to build a system that can per-ceive that."

For autonomous vehicles to eventually reach the road, they also must be able to do more than act as the driver's eyes. They must be the driver's hands and feet, as well.

Automakers have already begun to do that by severing the physical links that tie the driver to the tires, brakes and engine. The new breed of devices, called "drive-by-wire" systems, use the movements of the driver's hands and feet to send messages. And the trusty components that traditionally served as the driver's connection to the road act as the messengers.

Accelerometers help with navigation by providing signals for dead-reckoning calculations.

The steering wheel, for example, is transformed to a hand wheel. Like a joystick, it sends a command for the wheels to turn, but it does not do the turning.

Similarly, the brake pedal becomes a sensor. It isn't mechanically linked to the brakes. Instead, it sends a wee electrical current down a wire that tells the brakes what to do.

And the accelerator pedal? Another sensor. The touch of a foot is acted upon by a microprocessor, which commands the throttle to open or close.

At least one of those systems-throttle-by-wire-has already been implemented in millions of vehicles. Brake-by-wire and steer-by-wire will follow in the next seven years, engineers say, long before anyone dares to consider self-navigating vehicles. Such systems will be key to implementation of collision avoidance, which calls on the vehicle to take over the steering wheel and brakes.

To make by-wire systems serve in an autonomous vehicle, vehicle builders will need to take the technology one step farther. They will need to add systems that take the place of the human driver, and apply force to the steering wheel and brake pedal. Grand Challenge participants have accomplished that by employing electric motors.

SciAutonics, for example, employed three servomotors rated at 27 ft-lb each to work the brake pedal, accelerator pedal, and gearshift lever. It also used a large servomotor in conjunction with a belt drive to turn the steering wheel.

In the grand scheme of autonomous driving, by-wire systems will serve as the final link in a chain that begins with the radar sensors, which act as the driver's eyes, followed by the processors, which serve as the brain. By-wire systems will take commands from the "brain" and play the role of the driver's hands and feet, turning the wheel, and applying force to the brake and accelerator pedal.

Ultimately, engineers even foresee a day when vehicles will aid each other, sending wireless communications to one another when they are about to stop, approach an intersection, or steer around an obstacle in the roadway.

"Short-range communications could one day play an important safety role," Schulmeyer said. "The question is: How will we keep the attention of the driver when the vehicle is doing most of the driving?"

Despite all of the unanswered questions, many engineers believe that there will be a logical progression of technologies that will give drivers time to adjust as machines usurp more control. In the earliest stages, vehicles will provide audible warnings when drivers make mistakes.

Later, they will take control of the wheel and steer drivers out of trouble. Next, autonomous systems will take control in low-traffic situations. Finally, the vehicles will respond to user commands with door-to-door service.

"Twenty-five years is a very reasonable time frame to evolve through that continuum," said Hall of Velodyne Acoustics.

"Right now, we're still in the very early stages of autonomous vehicle control. But there's no reason we can't reach full door-to-door control eventually."