Former Apple engineers working on new eyes for driverless cars
Soroush Salehian raised both arms and spun in circles as if celebrating a touchdown.
Across the room, perched on a tripod, a small black device monitored this little dance and streamed it to a nearby laptop. Salehian appeared as a collection of tiny coloured dots, some red, some blue, some green. Each dot showed the precise distance to a particular point on his body, while the colours showed the speed of his movements. As his right arm spun forward, it turned blue. His left arm, spinning away, turned red.
“See how the arms are different?” said his business partner, Mina Rezk, pointing at the laptop. “It’s measuring different velocities.” Messrs. Salehian and Rezk are the founders of a new Silicon Valley startup called Aeva, and their small black device is designed for self-driving cars. The veterans of Apple’s secretive Special Projects Group aim to give these autonomous vehicles a more complete, detailed and reliable view of the world around them — something that is essential to their evolution.
Today’s driverless cars under development at companies like General Motors, Toyota, Uber and the Google spinoff Waymo track their surroundings using a wide variety of sensors, including cameras, radar, GPS antennas and lidar (short for “light detection and ranging”) devices that measure distances using pulses of light.
But there are gaps in the way these sensors operate, and combining their disparate streams of data is difficult. Aeva’s prototype — a breed of lidar that measures distances more accurately and also captures speed — aims to fill several of these sizable holes.
“I don’t even think of this as a new kind of lidar,” said Tarin Ziyaee, co-founder and chief technology officer at the selfdriving taxi start-up Voyage, who has seen the Aeva prototype. “It’s a whole different animal.” Founded in January and funded by the Silicon Valley venture capital firm Lux Capital, among others, Aeva joins a widespread effort to build more effective sensors for autonomous vehicles, a trend that extends from start-ups like Luminar, Echodyne and Metawave to established hardware makers like the German multinational Robert Bosch.
The company’s name, Aeva, is a play on “Eve,” the name of the robot in the Pixar movie “WALL-E.”
The market for autonomous vehicles will grow to $42 billion by 2025, according to research by the Boston Consulting Group. But for that to happen, the vehicles will need new and more powerful sensors. Today’s autonomous cars are ill prepared for high-speed driving, bad weather and other common situations.
The recent improvements in selfdriving cars coincided with the improvements offered by new lidar sensors from a Silicon Valley company called Velodyne. These sensors gave cars a way of measuring distances to nearby vehicles, pedestrians and other objects. They also provided Google and other companies with a way of mapping urban roadways in three dimensions, so that cars will know exactly where they are at any given moment — something GPS cannot always provide.
But these lidar sensors have additional shortcomings. They can gather information only about objects that are relatively close to them, which limits how fast the cars can travel. Their measurements aren’t always detailed enough to distinguish one object from another. And when multiple driverless cars are close together, their signals can become garbled.
Other devices can pick up some of slack. Cameras are a better way of identifying pedestrians and street signs, for example, and radar works over longer distances. That’s why today’s selfdriving cars track their surroundings through so many different sensors. But despite this wide array of hardware — which can cost hundreds of thousands of dollars per vehicle — even the best autonomous vehicles still have trouble in so many situations that humans can navigate with ease.