Autonomous cars: Sooner than we think?

Last week, the New England Motor Press Association (NEMPA) and MIT held a technology conference at the institute on the state of the art of autonomous vehicles. The experts there, along with presentations at last month's SAE Congress in Detroit, paint a picture of where the technology for self-driving vehicles stands—and how far we have to go for true autonomous operations. (Presentations from the NEMPA event should be available soon at the link above.)

One of the highlights of the SAE meeting was a presentation by Anthony Levandowski, business lead for Google's Self-Driving Car project. He noted that today "driving is the distraction" for many drivers who engage in other tasks while at the wheel. As part of the company's mission to foster innovation and meet challenges that will have a high, positive impact on society, developing technology for self-driving cars can reduce road fatalities and the more than 90% of collisions caused by human error. Google looks to furnish building blocks of such technology for industry.

Levandowski "drives" the Google vehicle to work about 60 miles one way. When he "boots up" the car, a digital world model, based on Google mapping, provides the car's location within ±5 cm, including "knowledge" of lanes, crosswalks, etc. Onboard sensors detect surrounding vehicles and objects for processing the traffic situation to determine a path to follow. He admits that construction zones challenge the mapping-based architecture because they can have "daily" changes to painted markings and safety-cone placement.

He says his commute is three quarters in the auto mode. He first started driving completely autonomously, but "certain situations need work," and now highway driving is mostly on auto. He uses his "downtime" in the car not to text but to plan out his day, and feels less stressed when arriving at work.

As for what Google has learned from the development, Levandowski notes a "time-to-collision" metric, where the vehicle travels at the same speed before preventative action is taken in an accident scenario, is less than 7.5 seconds—two seconds faster than a human driver. The robot car is also faster, better handling, and safer over a lap course and in many driving situations—which can help to remove skepticism in adopting the technology.

Interestingly, when questioned about the aftermarket for autonomous systems, Levandowski states it could be easier to bring these technologies to the public by aftermarket suppliers because their design cycles can be shorter than those of the auto OEMs.

Finally, he observes that the first autonomous applications will be limited to monitoring functions, keeping the driver in the loop rather than "being asleep in the back." Software updates can improve and expand functions, but Levandowski says a major challenge will be in handing off control of a vehicle to a driver in the event of failure. Depending on the mode of driving and what task the "driver" is engaged in when a failure develops, about upwards of 10 seconds may be needed for a person to get ready to drive after ascertaining the information needed to complete a safe transition.

At MIT
The NEMPA presentations reflected Google's experience, and added insight from the standpoint of academic researchers and automotive OEMs. Bryan Reimer, of the MIT AgeLab and New England University Transportation Center says automation will not decrease inattentive driving until the advent of totally autonomous vehicles, because drivers would not monitor such semi-autonomous systems well, becoming bored or distracted, until a more optimized way of connecting the driver to such systems is found.

BMW's Tom Baloga, U.S. VP of engineering, adds his company is concentrating on "highly automated driving" to make a good driver better. He cites development of an effective head-up display for making a car more comfortable to drive. Systems cost and liability issues will hold totally autonomous driving back, he states.

GM's Nady Boules, director of its Electrical & Controls Integration Research Laboratory, see autonomous vehicles coming in stages. First, integrated sensor technology will enable "360° safety," full-time around the car—more effective than just the periodic scan of the human eye. Such developments will lead to "cars that do not crash," and eventually to vehicles that drive themselves—the latter dependent on effective control algorithms (in addition to sensors) and non-technical concerns of government regulation, resolution of liability issues, and, not least, customer acceptance.

In fact, GM says it has enough advanced driver assistance systems in place now, that by 2016, with the addition of a lane centering function, it sees semi-autonomous operation under some conditions with Cadillac models.

Jonathan How, aerospace controls professor at MIT, observes that Google has probably advanced the push toward vehicle autonomy by five years. However, he has concerns about its map-based guidance in a complex and dynamic road environment, such as Boston's notorious street scene, with the need to handle "normal and off-normal behavior of other vehicles and pedestrians for classification and intent recognition." Other issues he voices are adverse weather operations and understanding the attention levels required for vehicle operators.

Continental's Director of Engineering Systems & Technology, Christian Schumacher, also sees 360° predictive safety systems in all vehicle types around the year 2020. These will first enable low-speed systems, for semi-automated driving, such as "traffic jam assist." Likewise, Sascha Simon, product manager for advanced product planning at Mercedes-Benz, adds, "The autonomous car for its own sake is not going to work. You need human monitoring and intervention. [Customers] want to buy a car to drive it. But in congestion, say, [automation] can be used to mitigate by taking over."

Finally, weighing in with the all important view from regulators on issues concerning vehicle automation, was David Strickland, administrator of the U.S. National Highway Traffic Safety Administration (NHTSA) speaking at the SAE Congress. He notes three concerns: Redundancy and system safety in the event of failure; integration of the driver into the vehicle; and, perhaps important when highly automated vehicles start to appear on the roads, integration of those vehicles into the overall transportation infrastructure.

Reflecting on the latter point, I can't help but bring up the "Law of Unintended Consequences" in the form of a recent blog posting I read. This blogger said he couldn't wait until autonomous vehicle technology becomes common—this way other cars will automatically move out of his way as he works his way through traffic.


(And here is another take on this autonomous vehicle stuff.)

Go to the Automotive Designline home page for the latest in automotive electronics design, technology, trends, products, and news. Also, get a weekly highlights update delivered directly to your inbox by signing up for our weekly automotive electronics newsletter here.