December 18, 1997

Take your hands off that car!

MAURY WRIGHT, TECHNICAL EDITOR

Hands-free computer-controlled vehicles promise to improve safety and reduce emissions, fuel consumption, and traffic congestion on our highways. Believe it or not, the long-envisioned highways of the future are beginning to emerge.

In 1995, the US government quietly introduced the National Automated Highway System Consortium (NAHSC), thereby formalizing an organized effort to automate our roadways. Comprising universities, automobile companies, and regional traffic agencies, the consortium has developed an impressive set of scenarios that should ultimately lead to hands-free operation of automobiles. This effort is no  pie-in-the-sky dream. Manufacturers have demonstrated these technologies on actual roadways. Moreover, the systems should improve safety as well as traffic congestion, pollution, and fuel-consumption problems.

"Automated Highway System" (AHS) is a broad label for a variety of techniques that seek to address widespread problems that current vehicles cause. AHS includes both in-car, computer-based control systems and, in some scenarios, additions to the highway infrastructure. In-car systems rely on vision-, magnetic-, and radar-based sense elements for input to computer-controlled steering, acceleration, and braking. Infrastructure enhancements include magnetic and visual cues placed along roadways.

At first glance, you may think that, even if hands-free driving were technically feasible, it wouldn't be cost-effective. Actually, however, adding to the infrastructure proves far cheaper than building more highways. AHS should eliminate the need for more highways because the systems should increase the capacity of highways. Moreover, car manufacturers are planning smarter cars and understand that they must do so without huge price premiums. Fortunately, car manufacturers can take advantage of many technologies that the PC industry spends time and money developing and that experience the rapidly falling prices inherent in the PC industry.

Evolutionary AHS deployment

Even without the formal NAHSC effort, you can expect an evolutionary roll-out of AHS technologies. In fact, Toyota unveiled just such an evolutionary technology at the NAHSC's Demo '97 exhibition in San Diego in August. The company detailed and demonstrated four near-term AHS technologies followed by long-term plans for true hands-free driving. In one demonstration, an in-car system monitors the car's position in traffic lanes and provides an audible warning whenever the car crosses a lane boundary. A second demo gently nudged the steering column when the vehicle departed from its lane.

The third demo showed off a significantly enhanced cruise-control feature. Using a radar sensor, the new cruise control automatically detects slower traffic. A car with the new system automatically slows or accelerates to adapt to other traffic. The fourth demonstration further enhanced the adaptive cruise control so that it will work in stop-and-go freeway jams. I experienced the stop-and-go operation, and as a California driver who regularly faces traffic jams, I found the system amazing. The equipped car creeps along behind another vehicle, and, although the driver must steer, the system eliminates the need for constant acceleration and braking. Urban drivers would likely pay a hefty premium for this feature.

Toyota plans to soon begin offering these short-term features on some of its Japanese-market car models. In longer term plans, the company is developing vision systems that automatically detect brake lights and avoid obstacles. The company also plans car-to-car radio communications. The communication system would allow a car that encounters an obstacle to automatically warn following cars of the problem.

Truckers use collision avoidance

Even more surprising than Toyota's plans, however, is the fact that many drivers unknowingly are the targets of collision-avoidance systems everyday. Eaton Vorad has been shipping its Collision Warning System to the truck and bus market for several years. The system uses forward-looking radar with a range of 350 ft to warn drivers of obstacles. The radar provides such warnings even in foggy, rainy, snowy, and dark environments. Eaton Vorad conducted a study of five companies' 473 fleet trucks and claims that the fleet had 1.61 accidents per 1 million miles before installing the system and 0.38 accidents per 1 million miles with the system in place--an accident-reduction rate of 76% during 47 million cumulative miles of driving. Eaton Vorad has also introduced an adaptive cruise-control system for passenger vehicles.

Still the most intriguing developments within the NAHSC community center on hands-free driving. The three most compelling demonstrations of such technology feature systems developed at Carnegie-Mellon University, Ohio State University, and the California Partners for Advanced Transit and Highways (PATH) partnership of the University of California--Berkeley and the California-based Caltrans agency (Sacramento, CA). All three feature vehicles with computer-controlled steering, acceleration, and braking. The vehicles can change lanes and pass other vehicles without driver assistance. And the three take significantly different approaches to the problem of hands-free driving.

These systems fall into one of two broad categories. Some are strictly for use with highway lanes dedicated to automated traffic flow. These lanes would include roadway enhancements to aid navigation and lane control. For example, Ohio State University demonstrated an aluminum tape it developed with 3M that is much like the yellow and white tapes highways use for lane demarcation. The aluminum tape, however, provides a perfect guide for in-car forward-looking radar. Manufacturers of such a technology would likely deploy it in dedicated lanes, such as those in high-occupancy-vehicle (HOV) corridors that have found wide use in urban areas countrywide.

Autonomous vehicle control

The second category of AHS works in a so-called nonsegregated lane, in which some vehicles might be standard cars and others might sport automated control systems. Vehicles targeted at this scenario rely strictly on in-car sensors to guide a vehicle along standard unaugmented roadways. 

At Demo '97, Carnegie-Mellon's Robotics Institute presented by far the most compelling example of autonomous vehicles for use in nonsegregated lanes. The demo fleet of "free-agent" vehicles included two city buses that Carnegie-Mellon developed with Houston's Metro traffic agency, a minivan, and two passenger sedans all of which were equipped for hands-free operation. The Carnegie-Mellon team first made headlines in 1995 when it took a minivan--in hands-free mode--from coast to coast. At Demo '97 the team demonstrated a scenario with buses and other vehicles. In this demo, the vehicles passed each other, waited for other passing traffic before changing lanes, and stopped for obstacles.

The Carnegie-Mellon vehicles rely on relatively standard and inexpensive control systems. A PentiumPro-based computer running the QNX RTOS handles all data acquisition and control. A camera on the backside of the rearview mirror provides the input for lane tracking. The camera continuously supplies pixels into a frame buffer, and the CPU samples the buffer 20 times/sec. Typically, the vision algorithm concentrates on only 2500 to 3000 pixels in areas of the image in which the algorithm expects lane-guidance cues. For example, the vision algorithm looks for the oil stain that's typically found along the center of freeway lanes. It also looks for white or yellow lane boundaries and even the seams that occur in pavement between a freeway lane and a shoulder.

The vehicles use forward-looking radar to avoid obstacles and other traffic, two side-looking radars on each side of the vehicle to monitor blind spots, and a rear-looking laser to monitor following traffic. You might wonder about the mixture of laser and radar sensors, but the choice between the two appears arbitrary. Radar works in more adverse weather conditions, whereas laser provides slightly more accurate range data. In reality, AHS vehicles of the future will probably use both sensor types in every direction for redundancy. In the Carnegie-Mellon system, the CPU samples all of these sensors 10 times/sec. It sends control signals 20 times/sec to the steering column, accelerator, and brakes.

My ride in the automated Metro bus easily matched the ride you get with a driver. Onboard monitors provided a visual demonstration of the lane-tracking algorithm showing absolute lane boundaries and a target zone within which the vehicle should remain. The system appeared to maintain perfect recognition of the boundaries throughout a seven-mile ride. 

Minimizing traffic spacing

The vision-based lane tracking system works with vehicle-to-vehicle spacing of only about 45 ft or more. Once vehicles get closer, the forward-looking camera can't see visual cues early enough to allow accurate steering at freeway speeds. According to traffic agencies, drivers on busy freeways today also achieve about 45-ft spacing.

AHS technology must close that spacing gap to deliver on the promise of increasing roadway capacity. Moreover, maximized side benefits such as reduced fuel consumption and emissions occur only when vehicles are close enough to create a drafting effect. You might think that drafting only pays off in stock-car racing, but the NAHSC has data that proves otherwise. Platoons of cars 3m apart and traveling at freeway speed realize a 5 to 15% fuel savings over individual cars. The savings escalate with even tighter vehicle spacing.

By automating all vehicles in an AHS lane, transportation engineers believe that they can eliminate stop-and-go transients. NAHSC reports claim that platoons of cars operating at 60 mph realize 47% fuel savings and generate 23% less hydrocarbon emissions relative to a car in stop-and-go traffic. These savings result strictly from constant-speed operation. The savings are less dramatic when you compare AHS cars with nonautomated cars traveling at 60 mph. But cars on urban freeways don't travel at a constant 60 mph during rush hour.

The participants at Demo '97 uncovered a challenge that they must overcome before constant-speed car platoons can become a reality. The first challenge is how to handle lane tracking with closely spaced cars. The aluminum radar-reflective tape developed by Ohio State and 3M has some promise for lane control but also has potential problems. For starters, the radar that detects the tape is still forward-looking and, therefore, subject to interference from other cars. Moreover, the tape is most effective when applied down the center line of a lane yet is also susceptible to damage over time as vehicles cross the tape line. Another type of magnetic tape from 3M, however, can be buried 2 in. under the surface of the pavement. The company claims the installed cost of the magnetic tape would be $3 to $4/ft, whereas the installed cost of the aluminum tape would be $2/ft.

Magnetic nails provide control

The California PATH team showed perhaps a better technology and demonstrated with eight-car platoons. The PATH system uses magnetic "nails" embedded vertically into the pavement along the center of a lane at a spacing of 4 ft. The 1-in.-diameter, 4-in. nails can be installed at a cost of $5 each, or $10,000 per mile. Although the cost seems high, it's several orders of magnitude less than typical new-lane construction.

Demo '97 presented Buick LeSabres (Figure 1) having a variety of sensors, including three three-axis, flux-gate magnetometers under both the front and the rear bumpers. The magnetometers allow the onboard computer to monitor not only lane position at all times but also the exact direction, or heading, of the car. A Pentium-class computer easily handles the lane-control algorithm. The PATH scenario aligned eight or fewer cars in a platoon with 21-ft spacing between cars, which would more than double the capacity of a typical freeway lane. Forward-looking radar provided the data to control spacing.

The PATH presentation brings up other issues, such as how a car negotiates to join or leave a platoon when entering or exiting a freeway. The PATH team developed an essentially wireless, car-to-car LAN with a proprietary protocol to handle communication between cars. The lead vehicle receives requests from cars that wish to join or leave a platoon and issues commands to instigate such actions. The radio communications also allow the lead car to transmit explicit warnings about upcoming obstacles or maneuvers.

The cars get more than just lane-control data from the magnetic nails embedded in the roadway. The PATH team has experimented with using the polarity of the nails to convey data. Each nail can convey 1 bit. The scheme will also require that some bits be set aside for ECC. The bit rate doesn't provide a lot of data, but it could provide location information, such as the current highway mile marker. The scheme could also warn the in-car computer of upcoming curves and potentially provide the distance to an upcoming exit.

Vision range finder

The PATH team believes it can further reduce spacing in car platoons. At Demo '97, PATH also worked with Honda on just such an effort. PATH researchers equipped a Honda Accord with dual videocameras to augment a forward-looking radar (Figure 2). Using stereoscopic techniques and an affine-transfer algorithm, the team demonstrated vision-based spacing control that's accurate to 3m. The prototype system required a relatively expensive Texas Instruments (Dallas) C40 DSP, and its developers must further refine the system to make it cost-effective. Still, the technique would again double the  amount of traffic that can use a freeway lane.

The downside of all the PATH work is the infrastructure concerns, which don't necessarily stem from the cost of implementation. Instead, a PATH-like scenario will require nationwide--if not worldwide--standardization of the lane-control system, the car-to-car wireless LAN, software protocols for car-to-car communications, and other items. It's a technology that's probably at least 10 years away.

Nevertheless, adaptive cruise-control systems will be widely available in short order. Many new cars support direct wire-line control of braking and acceleration systems, so adding the adaptive cruise control will work even in the after-market. Expect to see new cruise control within a year at prices of $1000 to $2000. Toyota plans to offer it for even less as an option on certain luxury cars.

You will also be able to buy lane-departure and collision-avoidance warning systems in the near future. These systems will cost about the same as a high-end radar detector and will likely be about the same size. It will probably be five years before anyone ships a system that manipulates the steering wheel. Moreover, the first such systems will simply nudge the wheel upon lane departure--to alert a sleepy driver, for example.

Unfortunately, technology itself probably won't delay widespread deployment of true hands-free technology. Instead, the companies that want to sell the systems must get government agency approvals. They will likely have to address insurance and liability issues. And some of the most compelling schemes must wait for standards to be developed and the infrastructure to be deployed.

| EDN Access | Feedback | Table of Contents |