Key Sensor for Self-Driving Cars is in Short Supply

The shortage is forcing some companies in the field to limit the number of self-driving cars they put on the road for test driving. Researchers are also affected. A person working with the University of Waterloo’s autonomous vehicles lab, for instance, says the lab couldn’t move forward because a five-year-old lidar it owns recently broke.

The main manufacturer of lidars for cars, Velodyne, told the school it would have to wait at least six months to get a 360-degree-view lidar. By sheer luck, last week the team was able to find another Canadian researcher to lend them his lidar, the person said, while it awaits the replacement device.

Right now, the intense demand is proving a boon for the 33-year-old company, based in Morgan Hill, California. As of last fall, Velodyne expected to more than triple its gross revenue to as much as $190 million in 2017, up from around $50 million in 2016, according to confidential data viewed by The Information. Velodyne projected revenue of $320 million in 2018. Profit figures and projections couldn’t be learned, but Velodyne President Mike Jellen said in an interview the company prided itself on its ability to operate profitably.

At the moment, Velodyne’s bottom line is dented by efforts underway to ramp up production quickly, Mr. Jellen said. The company shipped several thousand lidars last year and plans to ship more than 10,000 this year, thanks to a new factory the firm has opened 20 miles north, in San Jose, he said. By next year, it plans to be on pace to make a million lidars per year.

“We grew by 100% last quarter in terms of shipments and want to grow by 200% this quarter,” he said. “It’s quite the surge. We’re trying to deal with all the urgent requests.”

Lidar stands for “light detection and ranging.” It continuously sprays beams of light in many directions to help the car “see” what’s happening around it and plan the right path. A mid-range Velodyne device costs between $30,000 and $40,000 apiece, depending on the number of devices being ordered.

Velodyne raised $150 million last year from Ford and Baidu, the Chinese internet firm working on autonomous cars, to finance the increase in production. The companies made the investments as a way to guarantee supply for their future fleets, should they choose to buy Velodyne’s products. Ford plans to put fully autonomous cars on the road by 2021, and such components decisions must happen several years before car production. A Ford spokesman said the company’s agreement to use Velodyne lidars isn’t exclusive; it’s looking at other potential partners.

For now, though, customers are griping about the shortage. “That company has the weight of much of the industry on it today, and they are far from a high-volume manufacturing operation,” said one customer. “They are basically passing out sensors on a limited basis to preferred customers,” he added.

New Entrants

The danger for Velodyne is that the shortage could force customers to look elsewhere. While there are few other companies that sell lidars for cars, numerous competitors are trying to get into the field. They range from secretive Silicon Valley firms like Luminar Technologies to better-known ones like Quanergy and Innoviz. None of those three have shipped devices that are fit for cars. Ibeo, a traditional auto component supplier, has an available product but the company has a small share of the lidar market because its devices are too expensive for what they do, executives say.

One key lidar startup to watch closely: Cepton Technologies, which has numerous Velodyne engineers and has been operating in “stealth” mode for the better part of a year. Cepton is already shipping its lidars to customers, said a person with knowledge of the company. Cepton uses a frictionless vibrating mirror, as opposed to ball bearings, to scatter its lasers, this person said. Big auto component suppliers Continental and Valeo have announced they will ship their own lidars to automakers in a few years.

One former customer of Velodyne, Google, has developed its own lidar technology which its self-driving car unit—now called Waymo—uses. Waymo recently sued Uber, whose Otto subsidiary is made up of several former Google car engineers, for allegedly stealing Google’s lidar designs. Uber called the high-profile case “baseless.” (Beyond Ford, Baidu and several other carmakers including Volvo and Audi, Velodyne declined to confirm specific customers.)

“That company has the weight of much of the industry on it today, and they are far from a high-volume manufacturing operation.”

Virtually every other autonomous car and shuttle maker depends on Velodyne lidars for autonomous test vehicles, according to executives at many those customers, as do tech firms including Uber and Apple. At the moment, those fleets range from a handful to several dozen per company, but some are expected to be in the hundreds by this year.

All of the companies, Velodyne included, are aiming to bring down the retail cost of lidars to a few hundred dollars apiece, or less. A car would presumably need only a few of them to get adequate 360-degree coverage. (Long-range detection of objects is more critical in front of the car than behind it, so not all lidars on a car need to have the same range.)

Lidar Shortcomings

Lidars for cars are difficult to make because they require high-speed electronics, expertise in optics and mechanics, and firmware and software development to process the data. Those systems must be combined and survive a harsh driving environment in various weather conditions.

Traditional lidars, made by companies in Germany, Japan and elsewhere, don’t provide 360-degree coverage; that was Velodyne’s breakthrough, using ball bearings to spin the lasers around. But such a device isn’t considered to be reliable and rugged enough for a car that’s on the road for more than 10 years, because the ball bearings wear down due to friction. The new entrants are working on lidars that don’t have ball bearings. Mr. Jellen of Velodyne said his company’s “slice” lidars, which won’t have ball bearings, would be ready by 2018. A car would need three of them to get 360-degree coverage.

Lidar sensors alone aren’t enough for a car to drive autonomously. Lidars obtain “structural” information, such as the physical shape of objects and how far away they are from the vehicle, but they cannot see “textural” data such as lane markings, the color in a traffic light, and what direction a pedestrian is looking. That’s what cameras do, says Erez Dagan, a vice president at Mobileye, the leading seller of microchips that process images from cameras on cars. Lidars are also lacking when it comes to seeing small objects such as a tire on the road, or determining the relative velocity of the moving objects they detect—an area where radar sensors excel, Mr. Dagan said.

Most developers thus use lidars in combination with radar and cameras to create a high-definition “map” of the car’s environment, down to centimeters in detail, and predict how objects—people and cars—will move around it. It’s possible that lidars may be made less relevant if radars and cameras improve significantly. That’s a risky bet that Tesla is making as it develops autonomous vehicle software for its cars.

Unlikely Leader

Velodyne is an unlikely leader in sensors. David Hall founded Velodyne in 1983 as an acoustics company that designed subwoofers. In the early 2000s, Mr. Hall started manufacturing the subwoofers to China, leaving him with an empty facility in Morgan Hill. He has said in interviews that he heard about a U.S.-government sponsored “grand challenge” for researchers to compete to develop autonomous vehicles. He built his own stereo camera system for the first grand challenge in 2004. After Mr. Hall saw the limitations of camera data during the competition, he developed the first spinning lidar and entered the second grand challenge in 2005, according to Mr. Jellen. Two years later, other researchers were using his first device, which went on to retail for more than $100,000. Last year Velodyne released $7,000 to $8,000 puck-like lidars. The relatively low price further catalyzed self-driving research.

Until last year, Mr. Jellen said, the No. 1 market for lidars was mapping. The devices helped companies capture data such as the location of stoplights and other objects for navigation applications, for instance. Some lidar-based mapping data is now being collected to help autonomous vehicles understand, ahead of time, the general environment they’re about to drive in. This year, lidars for autonomous vehicles—for the purpose of “seeing” objects around them in real time, including pedestrians—will surpass those used for mapping, he said. (Drones are another emerging category.)

Mr. Jellen said lidar prices will fall to several hundred dollars apiece, and high-end versions may cost several thousand dollars. Mr. Jellen said he expects 100 million lidars to be shipped annually within the next 15 to 20 years, though initial growth will only be in developed markets with good enough road infrastructure for self-driving cars.

He said the current generation of devices aren’t perfect and there’s still much improvement to be made, in terms of recognizing dark objects and narrowing the physical gap between the lasers that the device beams out so that the data it captures can be higher-resolution.