Driving as we knew it
Sep 19, 2013
By Doug Annett
Fact 1: Depending on the source you want to quote, driver error contributes to 90-95% of fatal motor vehicle crashes. The other few percentage points are split between vehicle malfunction or driver incapacity. As the design and reliability of cars and tires have improved over the decades, fatal malfunctions are becoming rarer. Driver incapacity due to strokes, seizures and heart attacks are on the rise.
Let me entice fleet administrators with Fact 2. There will be a day when driver error will be eliminated. Autonomous vehicles will transport your sales and service representative to their destinations safely and on time, allowing them to use that pesky travel time to get serious work done.
The question is how to bridge the chasm between Fact 1 and Fact 2.
It turns out that there are hundreds of questions to be answered before Fact 2 becomes realized. In July 2013, The U.S. Transportation Research Board convened its Second Annual Workshop on Automated Vehicles, attended by 300 brilliant researchers of all kinds: engineers, psychologists, doctors, lawyers, and regulators. Their task was to question what is not known about how autonomous vehicles will transform our driving environment.
This broad ranging debate was managed through 10 workshops, each focused on a particular area of research. I participated in the Human Factors and Human-Machine Interface workshop because this was most appropriate for assessing the impact of autonomous vehicles on driver training.
There will be a day when driver error will be eliminated. Autonomous vehicles will transport your sales and service representatives to their destinations safely and on time, allowing them to use that pesky travel time to get serious work done.
Where we are now
For decades, engineers have been developing machines that operate remotely, from lunar rovers to drone aircraft. The dream of building the driverless car is still a long way off, meaning decades, but many of the sub-structures that will make this possible are already on today's cars. Proximity sensors and electric power steering enable cars to parallel park themselves with braking assistance from the driver. Today, many mid-range vehicles are loaded with sensors enabling lane-departure warnings, blind-spot warnings, adaptive cruise control, and various brake assist systems. Volvo has its City Safety System, which detects people and obstacles and automatically applies the brakes as a collision mitigation intervention at speeds up to 50 km/hr. Mercedes-Benz ups the ante by linking its Adaptive Cruise Control with steering assistance, allowing the car to autonomously follow the car ahead at speeds less than 60 km/hr on straights and through gentle curves.
So far this is not great news for fleet managers, since most vehicles on selector lists don't come near the sophistication level of those just mentioned. Executive level vehicles have some of these features, but the overall value and sophistication of a near luxury perk drives the selection, not particular features. Nevertheless, it only took a few years for self-parking technology introduced in 2007 by Lexus to migrate down the food chain to cars like the Ford Focus. Early adopters of any technology (iPod, iPhone, iPad, iAnything) are willing to pay outrageous prices to be first on the block with the latest, coolest tool, but it isn't long until cheaper versions flood the market.
It isn't up to human resources and fleet managers to enable this transition to automated cars, but we need to be prepared for these changes. At present, cars are only able to handle one or in some cases two of the processes. Adaptive Cruise Control can manage space-keeping behind another car, but it can't steer. Lane departure intervention systems can maintain a lane position but not change lanes.
Every major car manufacturer and suppliers of all kinds were in attendance at the workshop, some playing their cards close to the chest as to how far they are along the development process. It isn't essential to be the first to market with a viable technological breakthrough, but nobody wants to be left standing at the gate. Google has a running start, in that its systems rely on mapping technology, where real world information from the vehicle's sensors are compared to mapping data for location placement. Workshop delegates were taken for a short ride in a Google car, which drove autonomously on a local freeway, interacting with light traffic in a consistent, smooth and innocuous manner. Google seems confident that it will have a viable system on the market before other manufacturers, although everyone admits that their current systems have not attained the goal of collision avoidance in emergency situations, which is the gold standard for performance. A non-empirical question arises as to whether the machine needs to perform better than any driver, better than most drivers, or better than some drivers.
Currently, sharing attention between the driving task and electronic tasks dramatically affects driving safety. The point of automated cars is to enable and encourage hands-free driving so as to increase productivity. Although it sounds like heresy, the discussion as to the cost effectiveness for automated cars will occur within the decade. Factoring in the cost savings for cars that can safely park themselves may tip the scale sooner rather than later.
A Canadian expert's perspective
For a Canadian perspective, I spoke with Steve Waslander from the University of Waterloo. He is an associate professor, a specialist in robotics, but has turned his attention to automated cars.
Steve would have found himself exactly in his element at the TRD Workshop. He is confident that automated vehicles will be on the road, likely by 2020, at a level where the cars will be able to perform the dynamic functions encountered in a commuting context, leaving drivers free to perform other productive business-type tasks. Although temporarily disengaged from driving, they must remain available to take over the dynamic functions should that become necessary.
Waslander admits that there will infrastructure issues for reliability of sensor data. Some elements are fairly simple, such as clearly painted lines. Others, less so. Snow removal is more problematic for Canadian roads than those in Palo Alto, but even traditional road repair techniques such as crack filling can confuse sensors.
The human factors workshop developed research needs statements in four areas:
Customization of vehicle behaviour under automation. In testing currently available systems such as adaptive cruise control (ACC), manufacturers found that drivers wanted more flexibility to adjust the following gap so that it more closely mimicked their own driving behaviour. In short, the drivers wanted the ACC to tailgate closer. Real-world usage of ACC hasn't produced the tragic consequences predicted by folks like me.
Return to control protocols. There will be situations where the automated systems simply can't perform the task anymore, perhaps because the environment is too complex, communication and mapping limitations occur, sensors fail, or something as simple as snow covering the lines on the highway limits sensor reliability. In the initial stages of automation, resolving these issues will be relatively simple, like switching off cruise control. As automation assumes more and more driver control of routine functions, the methods of alerting drivers and timing the return to control become factors. Essentially, automated vehicles allow and assume the driver will be distracted by texting, reading or at an advanced level of development, even sleeping. A mandatory capability for fully autonomous vehicles would be the ability to safely pull over and park if the driver does not accept a request from the vehicle to take back manual control.
Abuse/misuse of autonomous vehicle systems. This is blue-sky territory, but we know that drivers have frustrated the best intentions of engineers and regulators in the past on issues such as seat belt interlocks and emission control devices. We don't yet know how drivers will abuse or misuse the systems, but manufacturers must consider preventive measures to limit possible cases. The Google test drivers report that other drivers harass them all the time, to see if the robot car has the smarts to handle stupidity.
Communication with the driver. This area of research questions what the driver needs to know about the vehicle's status in the performance of its automated functions and how this information is best communicated to the driver.
All of these topics are inter-related. A driver may misuse a particular level of automation by depending too heavily on its functions during inclement weather. Perhaps a driver has customized the performance of the car to enable getting to a meeting on time and writing the presentation while in transit. The driver may be too distracted to respond to car's request for returning dynamic control to the driver.
The key from Waslander's perspective is price. If an automated car can deliver the combination of safety, mobility and the perceived benefit of allowing attention-sharing at a reasonable price, he believes the market will accept it. Individual drivers and more recently fleets are accepting the higher price point of approximately $3,000 per unit for cleaner, greener and more fuel efficient hybrids and diesels. People pay almost this amount for the suite of electronic options which include a navigation system, satellite radio, and various connectivity interfaces. Exactly what the price point will be for automation is yet to be determined, but the market will settle this debate.
Counter to all the enthusiasm about the efficiency of an automated roadscape, Waslander presents a unique view. If drivers are able to work productively in the traffic jam, where is the motivation to improve transit or live closer to work? Would we actually tolerate even greater commute times, because we are using that time for work? Will drivers complain, "I'm almost finished this report. I hope this traffic jam lasts a bit longer."
Waslander would agree with TRB workshop participants that it is critical for the industry to get it right as to the market acceptance of automated cars. There are the classic failures: Edsel, New Coke and Microsoft Vista come to mind. In the automotive world, there is widespread acceptance that the introduction of Anti-Lock Braking Systems (ABS) was poorly handled. The assumption was that television ads depicting smiling people avoiding horrific crashes would do the trick. In fact there was a great deal of controversy over the efficacy of ABS in its early iterations. In conjunction with poor product knowledge in the showroom, customers had little understanding of what ABS could and could not do. Even 25 years after ABS hit the market, we meet drivers who have never been properly trained in ABS usage, as if the understanding would somehow be absorbed osmotically by drivers over the years.
Unfortunately, this approach may be at work already with automated cars. Gradually ever more advanced safety systems are being introduced into the market. The early adopters who understand and appreciate the technology of automation will accept and fully exploit each new component. The transition to full automation will be easy for them because they have been engaged with the technology at every step. For many other people, automation will seem to have burst onto the scene.
It isn't good enough to believe that "digital natives" (people born in the 21st century who never knew a non-digital world) will intuitively discern the issues of attention sharing. They will after all still have to drive the car in a Newtonian world at least some of the time. The issue of becoming too reliant on safety and convenience features has been talked about for years in the automotive world. Waslander spoke of research done by Volvo which showed drivers were very tolerant of fully automated cars and learned to trust their well-being to the technology remarkably quickly. The danger is in a partially automated stage where drivers learn to develop, as Waslander states, "trust beyond automation's capabilities."
My concern is that most drivers will not get the careful introduction to automation that research subjects do. The worst case scenario will be that the general population will learn how to use automated cars like they first learned to use ABS - by accident or perhaps by collision - unless a systematic approach to learning the technology is presented. It is not too early to think about this.
Perhaps next year, I won't be the only traffic safety educator at the workshop.
Doug Annett is director of operations for Skid Control School in Oakville, Ontario. Doug is a traffic safety consultant for business and delivers corporate training across the country through Skid Control School's national program, SafeRoads Canada. Doug can be reached at 905-827-5413 or by e-mail at firstname.lastname@example.org.