In a steamy basement garage in San Francisco’s Soma neighborhood, Cruise Automation CEO Kyle Vogt shows off the Generation 3, a car that looks like the electric Chevrolet Bolt, but which Vogt is calling the first production model self-driving car in the world. At least, it will be, once he and his team figure out the hard part.
Conquering self-driving car is about the car, sure. Predictions vary wildly, but it feels safe to say that at some point in the next 50 years, an automotive company will need to roll driverless-ready vehicles off production lines, and at a clip. That’s why Cruise and parent company General Motors showed off the Generation 3 on Monday, built in GM’s Orion assembly plant in Michigan. The vehicle comes with fully redundant systems, which means if one, say, computer that crunches data and determines how the car will drive dies, another stands ready to back it up—and keep human occupants safe from harm.
It’s an impressive hardware feat, but cracking autonomous vehicles remains a serious software game. Roboticists and engineers scramble to build computer systems that effectively combine the data streaming in from several different kinds of sensors mounted on the vehicle, sensors that help the vehicle “see.” A truly driverless car will have to process that information quickly, and know how to pilot through a variety of weird situations: if a cat jumps out into the road, or if it’s snowing, or if a vehicle isn’t allowed to turn right on red in this city, but is in the next one over. “The biggest stumbling block is going to be software at this point,” says Karl Iagnemma, the CEO of the self-driving startup NuTonomy.
Cruise knows that as well as anyone. At Monday’s event, Vogt read all of the steps to full self-driving off a small piece of paper. “You need safety and validation and automotive grade engineering; you need the actual AV software, which is the brain that drives the car; you also need simulation, testing, and validation facilities where you can test fleets, with lots of people to operate them; you need mapping systems, routing, supply and demand balancing or marketplace systems; you need to develop apps, APIs, in-car UX, data storage systems, analysis for all that data; you need activity solutions and then you’ve actually got to operate and maintain these cars once they’re out in the wild; you have operations and cleaning and maintenance and charging and construction, customer support, market operations,” says Vogt. He takes a breath. “And then you need high-volume vehicle manufacturing.” So, you know, baby steps.
Still, don’t discount the work it takes to build a self-driving car, even one that can’t quite self-drive yet. In normal vehicles, humans provide the failsafe—if your automatic steering system brakes, a fleshy arm can always reach out and redirect the column. But a truly driverless car won’t have a person up front, which means it must protect itself from mishaps. That’s the significance of the Generation 3 vehicle’s redundant mechanical systems, steering systems, braking systems, sensors systems, and computer systems. Once Cruise figures out how to self-drive, this car should self-drive safely—no conking out in the middle of the highway.
And while the Gen 3 may look like a Bolt, engineers had to modify 40 percent of the Bolt’s parts to accommodate autonomous driving. It should still maintain the Bolt’s impressive range, though: 238 miles on one charge.
General Motors says its capable of rolling out tens of thousands of Generation 3 cars per year, the way it does for other vehicle models. (GM sold 10.3 million cars in 2016). In other words: When and if Cruise cracks the computer part, GM can scale like bananas. They’re certainly pursuing their benchmarks aggressively so far; GM acquired Cruise just a year and a half ago.
They’re also not alone. In December, the Italian-American carmaker Fiat Chrysler Automotive delivered 100 Chrysler Pacificas with their own wiring harness, for specific use in Google’s Waymo self-driving fleet. Volvo, too, has production plans, promising a self-driving vehicle with full redundancy by the end of this year.
About 50 of the new GM vehicles should hit the streets in the next few weeks. (Local GM employees—but not members of the public—can hail these vehicles with an app, and use them to get around town. There will still be a driver inside, for safety). Expect to see other models as well—the Generation 1 vehicles that have cruised (sorry) around San Francisco for the last 14 months, and then about 130 Generation 2s, “hand-built, artisan-designed” vehicles with nearly double the sensors at the Gen 1, but without the large-scale production capabilities of the Gen 3.
GM won’t commit to a deadline, but once its vehicles are fully ready, don’t expect to find them on showroom floors. The technology will be so expensive that the company will deploy them in fleets, picking up passengers like taxis. Or Lyfts—the Detroit automaker struck a deal with the ridehail company last year to run its tech on their platform.
Besides, all of that’s still a long. “The big thing holding us back at this point is the software,” says Vogt. “We still have a long way to go before we reach that bar.”
For now, the future self-driving car looks like, well, a car, with a steering wheel and gas pedal demonstrating that humans are still part of the equation. But when cars go full-on driverless, those will be a liability—you don’t want pesky people messing with your fancy systems. There are plans to rip those out, GM says. Eventually.