Luminar, a lidar startup founded by a 16-year-old, has come of age. Founder Austin Russell is now 23, and he tells IEEE Spectrum that his company has started mass production.
“This year we will produce 5000 units per quarter, enough to equip every autonomous car unit on the road,” he said, during a visit to our offices in Manhattan on Monday. “We had been using optics PhDs to hand-assemble them; by year’s end one’ll be coming off the line every 8 minutes.”
That’s an achievement, and Russell has evidently kept the production milestone under his hat for a while. After all, we’re 12 days into the second quarter. Russell is good at staying under the radar, having kept Luminar in stealth mode for its first five years. And, at 6 feet, 4 inches, Russell has to work at being inconspicuous.
Lidar startups have come up like mushrooms—Russell says he’s tracking about 60 of them—but up to now none of them could match production numbers with the industry’s pioneer, Velodyne, whose roof-mounted rotating beacon has become the identifying mark of the self-driving car. Many of those startups haven’t even got a working prototype to show off.
Six months ago Toyota announced that its experimental self-driving car was using Luminar's lidar. "Three other major auto makers have committed to using our platform for all their development," Russell says. "These fleets will ultimately evolve into serial production models for the market."
Luminar’s set can see 250 meters down the road, 50 m more than the prototype could manage a year ago. That would provide plenty of time for a self-driving car to react to events, even at highway speeds.
The Uber taxi that killed a pedestrian last month was apparently doing 60 kilometers per hour (38 mph). At that speed, Russell says, his company’s lidar can give about 14 seconds’ warning. To be sure, it’s not yet clear whether the Uber accident had anything to do with the lidar’s capabilities.
When I last met up with Russell, a year ago, he had one of the handful of prototypes his company had made under his arm. It was a rather substantial machine, but today’s production version is substantially smaller and lighter.
It works by splitting a laser beam into two parts, one for each of two windows, placed side-by-side at an angle. Each covers a 60-degree field, for a total of 120 degrees; to give a car full, 360-degree coverage with a certain amount of redundancy, you’d probably want four units. Microscopic moving mirrors steer the beam through the field of coverage. Paired with the laser is a photoreceptor that scans the reflected light.
The default setting is to focus on the horizon, at 10 frames per second. It can scan as slowly as 1 frame per second for “absurdly high resolution,” Russell says, or go up to 20 Hertz, for equally absurd quickness, at the cost of resolution. The tradeoff can be specified by the car’s software.
But the real trick is simple, brute-force laser power. Luminar can shine its single pencil of light brighter than others do because it works at a wavelength of 1550 nanometers, which hardly can penetrate to the retina. Our own Evan Ackerman took a deep dive into the optics back in July.
The wavelength can be handled only by the compound semiconductor indium gallium arsenide, a costly material. Up until now only the military could afford to work with it. But Luminar’s chief talking point has been that the lidar feature that car makers most need to optimize is performance, not price. That’s also been the implicit position of industry pioneer Velodyne, which has charged upwards of US $70,000 for its most capable models. Tellingly Velodyne has not yet specified the price it’ll ask for its newest and most capable lidar, the VLS-128.
“A 3-inch array of inGaAs used to cost $30,000,” Russell says. “But we use very little—about the width of a human hair. It costs us $3 to build the entire receiver module. Of course, what it costs us to make is not the same as what we charge for it!”