Driverless Cars Are Data Guzzlers The Self-Driving Car of the Future Will Consume an Enormous Amount of Information

Driverless Cars Are Data Guzzlers

The Self-Driving Car of the Future Will Consume an Enormous Amount of Information

UCILIA WANG

Updated March 23, 2014 4:36 p.m. ET

If the car of the future is going to be able to drive itself, its appetite for data will be almost insatiable.

That’s because to drive themselves safely and reliably, cars will need numerous sensors to gather as much information as possible about their constantly changing surroundings, and powerful onboard processors to analyze that information. Some data may flow to remote computers for crunching, too. Cars will need to talk to each other, as well, for a better sense of how traffic patterns in the immediate area are developing. And all of this will have to happen lightning fast.

“Building a robot that can understand everything that is going on around it—knowing exactly where the stop signs are, decoding the lanes around it and never mistaking mailboxes for pedestrians on the side of the street—is really hard,” says Edwin Olson, a University of Michigan professor who is working with Ford MotorCo. F -0.51% on autonomous-car research. “The variety of odd situations that a vehicle can find itself in is so high.”

Here’s a closer look at what’s involved in making all this happen.

Gathering Data

It all starts with a clear picture of a car’s immediate surroundings as it zips around. Car makers are experimenting with different types of sensors and different combinations to figure out the best way to detect moving and stationary objects and gauge their relations to the car and to one another.

A laser-based sensing technology that excels at creating high-resolution, three-dimensional images, called lidar, is a popular choice. Ford, for instance, mounts four of these sensors on the roof of its research car. Collectively, they can capture 2.5 million 3-D points per second within a 200-foot range.

The Audi NSU.XE -0.01% unit ofVolkswagen AG VOW.XE -1.58% places a lidar sensor on the front bumper of its experimental self-driving car, and radar sensors and cameras around the body to complete a 360-degree view.

Cameras are especially useful for detecting changes in traffic lights and helping with navigation, and they can gather a lot of data—between 20 and 60 megabytes per second.

Sonar sensors also are being used to detect objects and calculate their distance, direction and speed. And all the while, GPS receivers pinpoint where the car is and where it’s heading on the map.

Split-Second Analysis

Here’s a software challenge for the ages: All that data needs to be analyzed quickly enough so that the vehicle can react to changes in its surroundings in less than a second. (A very good human driver takes less than one second to react; the typical driver can take three to six seconds.)

Consider some of the likelier scenarios, not to mention the unexpected: The car’s brain has to measure variables including the vehicle’s speed and road conditions, in addition to analyzing the surrounding traffic, to figure out how hard and fast to brake when necessary. It has to gauge the flow of traffic to merge on a freeway. On city streets, it has to analyze the often erratic behavior of a multitude of pedestrians, bicyclists and other cars.

Computing Power

Researchers say processors today have ample firepower to crunch the necessary data. The key is to design the computing modules to tuck neatly into a car.

Information collected by sensors in cars today flows via wires to different sets of chips in the vehicles for processing. This distributed-computing approach will likely remain as cars become more automated, but how many sets of computing modules will be needed is an open question, says Greg Stevens, global manager for driver assistance and active safety at Ford. Audi is working on a more centralized computing approach, and managed to shrink its driverless car’s main computer, which filled the trunk about a year ago, to the size of an iPad, says Bjorn Giesler, project leader for piloted driving at Audi.

Security and Privacy

Any flow of information between cars or between a car and a remote computer will be vulnerable to security breaches intended to steal data or to disrupt cars’ ability to navigate and make good decisions. Hackers could track cars’ movements and possibly even take over their operation. If millions of driverless cars are on the road, the challenge of protecting data on that scale will be enormous.