Toyota and a Boy Wonder Team Up on Self-Driving Cars – WIRED
Renting out a 100,000-square-foot San Francisco cruise ship terminal feels ostentatious, even for a Silicon Valley coming out party. But Luminar, a small company dedicated to a far corner of the nascent autonomous vehicle industry, and its 22-year-old founder, Austin Russell, happily shelled out thousands of dollars to rent Pier 35. Not for a fun or lavish celebration—in fact, there are only about six of us gathered in the cavernous building—but because it’s one of the few places in the city big enough to demonstrate the power of the lidar system Russell has spent the last five years creating. A system that Toyota, a major carmaker with rapidly expanding self-driving car ambitions, says it plans to adopt for use in its vehicles.
Standing at one end of the empty building, Russell hands me a pair of binoculars and points my gaze to a 30-by-30-ish square of black posterboard, resting on an easel 200 meters away. This is no ordinary poster board. This material, called Permaflect, is specially certified for “uniform spectral response,” to reflect just 10 percent of light that it hits (and it costs between $10,000 and $20,000). It’s the official tool for testing lidar sensors, the radar-like devices that build an image of the world by firing lasers beams and measuring how long each takes to bounce back after hitting an object.
Russell shows me the readout from his lidar, projected onto a large monitor. In this world view of bold colors, the black square glows orange. The other targets Luminar has set up along the way—a pile of tires, a mannequin, a parked Tesla Model S and BMW i3—pop just as clearly. I can make out the individual feet of the pigeon who has joined the party, strutting back and forth.
“Autonomous cars can’t reliably see today,” Russell says. “We need fundamentally better hardware, fundamentally better data.” That’s what Luminar, which came out of stealth in April, now offers. Its shoebox-sized lidar scanner, Russell says, sees 10 times farther than its closest competitor, with 50 times better resolution.
Russell’s entry into the market puts him firmly in the race with numerous other companies racing to build fully autonomous cars. The engineers behind this technology believe lidar is a crucial to the future of AVs, and dozens of companies, small and large, are trying to build a lidar unit with the perfect balance of range, resolution, manufacturability, robustness, and cost. (Lidar intellectual property is at the center of the heavyweight legal brawl between Google and Uber.) Lidar is more precise than radar and doesn’t depend on ambient light, as cameras do.
Just a few problems with the technology: Compared to cameras and radar, lidar is very nascent. Today’s hardware isn’t much for life on a moving vehicle, exposed to the elements and bumps in the road. There’s also a major naysayer in the business: Elon Musk, the CEO of Tesla, thinks lidar will always cost too much to build into consumer cars. He’s betting that his cars can safely navigate roads by relying on good cameras and artificial intelligence.
“Sure, maybe 100 years from now,” Russell retorts. He insists lidar is the marquee sensor, the way to give the car’s software the quality data it needs to make driving decisions in any situation. Something of a wunderkind, Russell spotted this market opportunity six years ago. At 17, he dropped out of Stanford a few months into his freshman year, and with funding from a Thiel Fellowship, started Luminar. He created his own hardware solution, changing the frequency at which the scanner emits light (from 905 to 1550 nanometers), and ditching a silicon receiver for a more expensive one made of indium gallium arsenide. The result, Russell says, is a lidar that’s both more capable and more costly. He’s not too worried about the (unrevealed) price tag, saying “relevant players” in the industry have told him they’d gladly pay $400,000 per car for a perfectly safe self-driving system.1 That price tag would certainly make the tech unacceptable for personal cars, but could be more easily amortized on fleet vehicles in taxi services that run nearly constantly.
Russell’s early bet seems to be paying off. Luminar now employs 250 people, split between Orlando and Portola Valley, to the southwest of Palo Alto. This spring, it started building its hardware at scale, starting with a production run of 100 units. The next run will be for 10,000 units.
And now, Toyota is throwing its weight behind the startup, announcing today that its latest AV, the inspirationally named Platform 2.1, will use Luminar’s lidar to see the world. The world’s largest automaker didn’t say much about autonomous driving until late 2015, but when it did, it made quite a statement: It pledged to spend $1 billion on artificial intelligence and robotics, and opened research labs in Silicon Valley, Cambridge, Massachusetts, and Ann Arbor, Michigan.
“Perception is one of the keys,” says James Kuffner, the CTO of the Toyota Research Institute. (Kuffner was also a founding member of Google’s self-driving car project, and led that company’s robotics division for years.) “Lidar is one of the critical sensing modalities, because it gives you detailed geometry at range.” Quality data, in other words, that Toyota’s software can then analyze. “We’re evaluating all kinds of prototypes,” Kuffner says, but Luminar has a novel approach, and has proved it hits that 200 meter range mark, considered a vital requirement. (A Toyota spokesperson declined to reveal how many units it’s buying from Luminar, or what it’s paying for them.)
Russell hasn’t won the lidar wars just yet, but the deal with Toyota is something to celebrate. Maybe even in 100,000-square-foot cruise terminal.
1Story updated 12:33 ET on Wednesday September 27 to clarify theoretical cost of a fully autonomous car.