The trail to the future
The trail to the autonomous car’s future, however, might not be as clear as the Google Car seems to blaze for companies in the automotive industry. While Google is charting its course in autonomous car development through the self-contained vehicle, others are opting for a future in which the car’s sensor data is connected with information coming from the outside world.
For example, HERE (built on more than 25 years of experience in cartography), a Nokia business, has partnered since last fall with the Mercedes-Benz division of Daimler and two automotive system integrators, Continental Corporation and Magneti Marelli. Their goal is to develop jointly smart maps for connected cars and ultimately self-driving cars, Lanctot explains.
The third option for carmakers is to wait for the vehicle-to-vehicle infrastructure to get ready, he says. But by 2020, when the mobile telecommunication standard starts moving to 5G, where latency is expected to become much lower, “it would pretty much spell the end of V2V,” he predicts. There are many other challenges to overcome before the automotive industry rolls out an autonomous roadster.
The industry needs to build an autonomous car that can handle a broad range of circumstances -- from driving on high-speed highways to navigating low-speed surface streets, he says. On a freeway, for example, an autonomous car is expected to take over while you pick up a pair of sunglasses dropped on the floor, says Lanctot. On surface streets, the same autonomous car might have to execute a sudden stop when a ball rolls into its path, he added. Naturally, it needs to self-park, probably not such a big deal.
One city at a time?
Google’s Urmson wrote in his blog: “We still have lots of problems to solve, including teaching the car to drive more streets in Mountain View before we tackle another town.”
Hang on. Does that mean that Google Car needs to learn to drive one city at a time?
IHS analyst Juliussen says, “I am sure that the Google Car will take whatever knowledge it has accumulated in one city to another, but it’ll have to learn to drive different terrains -- a city, for example, that is very hilly.” Juliussen is impressed with what appears to be the judicious and meticulous approach Google is taking to Google Car’s machine learning. “It’s doing more testing, more proving and more learning in software,” he notes.
While acknowledging that self-driving cars are probably much more needed in the developing world, where car accidents happen more often, Google is taking Google Car to Western countries first, he says. “It’s obviously easier to drive and the risk is minimum.”
Next page: Forecasting "when"