Google's Waymo Sues Uber Over Lidar Self-Driving Car Technology

The biggest challenge in building an autonomous vehicle is giving the car the ability to see the world. It requires a thorough understanding of lidar, the radar-like system of lasers that creates the digital map each car needs to navigate the world safely and competently.

Mastering lidar is essential to the technological and commercial success of robo-cars, yet there is much work to be done. Most automakers are still figuring out how to make it robust enough for automobiles, and cheap enough for consumers. Doing this demands a serious investment and expertise. Waymo, Google’s self-driving car outfit, says hundreds of its engineers spent thousands of hours and millions of dollars perfecting the company’s lidar. And it accuses Uber of stealing its work.

In a lawsuit filed Thursday, Waymo claims former Google employee Anthony Levandowski downloaded 14,000 technical files from a company server, then used the information to launch the autonomous truck startup Otto. Uber acquired Otto a few months later and tapped Levandowski to lead its robo-car program.

“Otto and Uber have taken Waymo’s intellectual property so that they could avoid incurring the risk, time, and expense of independently developing their own technology,” Waymo says in its complaint. In a statement, Uber called Waymo’s claims “a baseless attempt to slow down a competitor.”

Lidar is an acronym for light detection and ranging, and just about everyone in the autonomous vehicle business uses it. (Tesla is an exception; Elon Musk says cheaper cameras and radar can do the job.) Simply put, lidar maps the world by firing millions of laser beams every second and measuring how long it takes them to bounce off nearby objects. That data creates a 3-D “map” of the area around the car. If you want a fully autonomous car, “you need lidar,” says Glen De Vos, CTO at the auto industry supplier Delphi.

Using the “point cloud” the lasers produce, autonomous driving software can pick out cyclists, pedestrians, and other vehicles. It can compare that data to a reference map of the area (also made with lidar), to see what might have changed (like a new lane or traffic signal). Such data lets the car infer its position with greater accuracy than commercial GPS. Unlike cameras, it works at night and it offers far better resolution than radar. “It’s a very good sensor in terms of range, distance, and resolution,” De Vos says.

Plunking a lidar sensor on your car doesn’t cut it. To do all the fun stuff—perception, navigation, localization—you need serious software expertise. Part of the work is configuring the sensors so they fire their lasers in the right way. This varies with the type of car, where the lidar units are on that car, and whether they are focused on short- or long-range sensing. You also need the algorithms that turn the millions of data points received every second into a point cloud.

Using more than one lidar sensor—Google uses three, Uber seven, Ford four—you must combine the data from each of them into a big picture, accounting for the position of each and the movement of the car. “It involves a lot of math and science,” says Anuj Gupta, a product manager at Civil Maps, a startup that turns lidar data into maps for autonomous cars.

It’s the most important piece of IP one can have. Louay Eldada, Quanergy CEO

Once you’ve got your point cloud, you learn to see where it differs from your base map. You don’t just spot and classify the important stuff like cyclists and other vehicles (while ignoring the leaves on trees and flying plastic bags), you track them as they move. And you keep doing it, every millisecond, as a staggering amount of information pours in. A lidar map of Palo Alto, California covering 300 miles of traffic lanes takes up one terabyte.

Doing this takes time, money, and specialized skill. Doing it on a commercial scale requires more of the same. Delphi took a shortcut in 2014 when it bought autonomous vehicle software startup Ottomatika. Dozens of engineers at the company, which spun off from work at Carnegie Mellon University, spent more than a decade on the system. Four months after acquiring the startup, a team of Delphi engineers rode cross-country in an autonomous car, underscoring the technology’s value. It can make or break an autonomous vehicle program.

“It’s the most important piece of IP one can have,” says Louay Eldada, CEO of lidar manufacturer Quanergy. “That’s your differentiator.” Waymo says Uber’s “differentiator” looks an awful lot like the one it built.