If you’re in Mountain View and you spot a self-driving car, the smart money says it’s one of Waymo’s modified Chrysler Pacifica minivans. If you see an electric car, it’s most likely a Tesla. But the odds on both are set to change. Waymo is rolling out its new fleet of roboticized, all-electric Jaguar I-Pace SUVs, a robotic armada that over the next two years will grow to as many as 20,000.
At first glance, the newly outfitted I-Paces don’t look much different from the mocked-up version Waymo showed off nearly two years ago, when it announced a partnership with Jaguar. A closer look reveals a bigger shift offering hints about the future of autonomous vehicles. The two rooftop lidars on the Pacificas have been compressed into one unit that can handle both medium- and long-range sensing. The perimeter lidars near the wheels have been joined by “perimeter cameras.” A lidar and camera sit on the grille, where you’d normally see Jaguar’s roaring cat logo. And so on.
To borrow an auto industry phrase, this is more than a “facelift,” a package of minor changes designed to refresh a stale model. Over the past several years, hundreds of Waymo engineers have rebuilt most of the company’s self-driving hardware, chiefly the cameras, lidars, and radars that perceive the world around the car. They did almost all of the work in-house and from scratch. Now the electric Jaguar (which Waymo, inexplicably, declined to dub the I, Robot-Pace) is the first vehicle to benefit from the company’s fifth generation hardware suite.
All told, the vehicle can now see farther and in more detail, says Satish Jeyachandran, Waymo’s hardware lead. The cameras, for example, now collect useful data from more than 500 meters out. All the sensors can see through and survive in nasty weather. The system can be readily reconfigured to work as well on Waymo’s 18-wheeler as on the I-Pace. Just as important, Waymo says it can produce the components in adequate numbers for a commercial self-driving fleet. “We wanted this generation to be much more scalable and manufacturable and reduce the costs of the system by half,” Jeyachandran says. “I'm super confident that we are pretty close to achieving all of these metrics.”
The new toolkit is the result of a long running, if quiet, push by the Alphabet subsidiary to invent its own hardware. When Waymo started life in 2009 as Google’s Project Chauffeur, it stuck off-the-shelf sensors on Toyota Priuses purchased at a local dealership. That included a roughly $80,000 rooftop lidar from Velodyne, which dominated the early laser scanner market. Soon, the Googlers developed their own lidar, which debuted around 2012 on their self-driving Lexus SUVs. When they moved on to the Firefly—the koala-like prototype without a steering wheel or pedals—they made much more of the hardware in-house, including the work of designing the cutesy two-seater.
Waymo (as it was christened in 2016) soon gave up building its own vehicles, but kept making the bits that replace the human driver, chiefly the sensors and the computer systems. The software may do the thinking, but it’s no good if it can’t rely on good sensing data, or reliable connections between the sensors and computers. The best way to ensure that everything going into the car meets Waymo’s standards, Jeyachandran says, is to do it themselves. “We realized that off-the-shelf doesn’t meet the requirements needed for self-driving.” With the financial backing of Alphabet and a head start in the field, Waymo could make the equipment it needed to see and navigate the world. In a statement, Velodyne says that it “continue[s] to develop sensor solutions.”
The effort—including paying hundreds of engineers for years of work—is not cheap. Nor is the output. Numbers are scarce here: In 2012, the company said its vehicles required about $150,000 worth of specialized equipment. In 2017, Waymo reported it had dropped the price of its lidar by more than 90 percent from the prior year. So the drive to reduce costs—especially for vehicles meant for the not-necessarily-lucrative taxi and delivery business—is easy to understand. This week, Waymo announced its first external investment, a $2.25 billion round led by Silver Lake, the Canada Pension Plan Investment Board, and the Mubadala Investment Company. Until this round, Waymo was funded by its parent company, Alphabet.
Waymo’s hardly alone here. The big players all run hardware programs. That’s largely because the self-driving industry is young and small enough that the supply base isn’t really there. A few years ago, there was no demand for a lidar that could spot a pedestrian wearing black, in the rain, at 250 meters. Especially not one that could handle life on the road, including the abuse of extreme temperatures, potholes, and being pelted with gravel and salt. So, many self-driving companies have moved to make their own goods wherever necessary, and possible.
“They’re all trying to build their own hardware,” says Kevin Peterson, CEO of the sidewalk delivery robotics company Marble, who worked on the Darpa Grand Challenges with many of the industry’s current leaders. “They’ve got the money, and they’ve got these needs that self-driving cars have and nobody else does.”
Cruise, Argo, and Aurora have all acquired lidar startups and are also developing custom lidars, and using off-the-shelf versions in the meantime. But Waymo had a big head start; it has already gone through multiple generations of hardware and developed several types of lidar for different uses. One big change with the I-Pace sensor suite is the rooftop lidar, which handles both long- and mid-range scanning (the Pacifica uses separate lidars for different distances). Lidar lead Simon Verghese declined to divulge specs, but says the resolution on the new unit is 10 times better than its predecessor, thanks to “new architecture.” And it’s just one more step forward. “I doubt this is the last lidar we build,” he says.
How a chaotic skunkworks race in the desert launched what's poised to be a runaway global industry.
Because it delivers detailed information that doesn’t require interpretation, lidar is the marquee self-driving sensor. But a self-driving car also relies on cameras (which offer better range and resolution, provided you can translate 2D images into 3D understanding) and radars (which see in all weather and can tell you how fast things are moving). So Waymo made its own versions of those, too. The 29 cameras on the I-Pace can handle a wider range of lighting conditions and better handle extreme temperatures. Engineers also created new “perimeter cameras” that sit near the wheels, to spot things very close to the car. Because they need to see in the dark and regulations bar the use of white lights anywhere but on the front of a vehicle, Waymo engineers added near-infrared lights to assist the cameras.
The radars, too, got a makeover. The systems powering adaptive cruise control and semi-autonomous driving in today’s commercial vehicles offer minimal detail. So Waymo built a new version, making its own radome, antenna, circuit board, mechanical housings, firmware, software, and more. “Everything on that was designed in-house, specifically for self-driving,” says radar lead Matt Markel. Where a standard car radar tells the computer, “object x meters ahead moving at y mph,” Waymo’s tech provides the resolution to distinguish a car from a truck and a motorcyclist from a pedestrian.
To protect their home-built computers in the trunk, Waymo engineers used a rubber mat that forms a watertight seal and is easy to clean—because who knows what might leak out of a passenger’s suitcase on the way to the airport someday. The ring of 16 cameras sitting below the main lidar each get their own wiper to clear dirt or rain, pulling fluid from an extra tank in the trunk.
And before any of it could hit the road, it had to endure savage testing mimicking the roughness of life on the road. In a new and lightly regulated industry, that meant getting creative. Airgun-wielding Waymo workers fired ball bearings at the shiny new cameras and lidars. They blasted radars with a firehose-level stream of water. They drove their cars over rumble strips and fields of gravel, subjected them to the heat of Death Valley and winter in Michigan.
Now that the first phase of closed course and durability testing is done, the team is moving the result onto public streets, first in Silicon Valley and then into its ridership program in the Phoenix suburbs. The hardware will then be reconfigured to ride on Waymo’s fleet of semi-trucks. And then the engineers will go back into their labs and start cooking up the next generation of the tech that will take their colleagues’ software into the world.
Updated, 3-4-20, 4:15pmET: An earlier version of this story incorrectly said Waymo's Jaguar I-Pace fleet would reach 20,000 over the next two years. The story also gave an incorrect amount for the cost of Waymo's lidar unit in 2017.