Will the Tesla Model 3 Be the First Truly Self-Driving Car?

Elon Musk has said that people should be able to summon their cars from across the country by the beginning of 2018, which happens to coincide with the Tesla Model 3's planned release date. Credit Photograph Courtesy Tesla

On the evening of March 31st, Elon Musk unveiled Tesla's sinuous Model 3, the company's first "affordable" electric-car model. After touting the sedan's punchy acceleration, two-hundred-and-fifteen-mile battery range, and sweeping, seamless glass roof, he mentioned its base price of thirty-five thousand dollars and told the audience that prospective buyers had already reserved more than a hundred and fifteen thousand of the vehicles, to rapturous applause and shouts of "You did it!" Not one to miss a marketing trick, Musk capped the night on Twitter, with a cryptic thank-you message that promised more: "Thanks for tuning in to the Model 3 unveil Part 1! Part 2 is super next level, but that's for later . . . ."

Within hours, the tech community was awash in speculation about what more Tesla could have in store for the Model 3. Some wondered, specifically, whether it would be the world's first mass-market, fully autonomous self-driving car. Spurred forward by Google and other Silicon Valley companies, the auto industry has been tinkering with autonomous vehicles for years. Tesla has demonstrated a unique appetite for risk, though, by equipping two of its cars, the Model S and Model X, with rudimentary self-driving capabilities. Musk has also said that people should be able to summon their cars from across the country by the beginning of 2018, which happens to coincide with the Model 3's planned release date. And, after the announcement, Bloomberg's Tom Randall noted that Musk kept referring to the Model 3's "steering system" or "steering controls," rather than its steering wheel, and that he'd said that the system shown at the Model 3 lau nch wasn't final. But perhaps the most compelling evidence that Tesla's Model 3 may have significant autonomous capabilities lies in the company's unique technological approach, which could allow it to achieve a "hands off" driving experience at a fraction of the cost of its competitors—including Google, the heavyweight in the field. (A disclosure: one of the authors, Levi Tillemann, owns stock in Tesla.)

To understand why Tesla's strategy may give it a decisive advantage, it helps to know the history of its main competitor's approach. As Burkhard Bilger detailed in this magazine in 2013, the research began with a small group of Bay Area roboticists, led by Google's Anthony Levandowski and Sebastian Thrun, who bought a Prius, rejiggered its electronic controls, and installed a laser-based system, known as lidar (for light detection and ranging), that measures the physical distance between the car and the objects around it. At first, they worked on the vehicle as a side project. But then Google brought the endeavor in-house and gave it a basically unlimited budget.

There's a very good reason Google and other manufacturers have bet on lidar: by repeatedly firing pulses of laser light in all directions and timing their return, lidar can create a precise three-dimensional map of the car's surroundings, accurate down to a couple of centimetres. This precision allowed the Google car to progress much faster than most people expected. (Google demonstrated these advances in a 2012 YouTube video featuring Steve Mahan, who is legally blind, as he went about his daily routine in an autonomous Google car.)

But while much of the tech community marvelled at Google's achievement, Musk thought that lidar was too expensive. (A top-notch system costs about eighty thousand dollars.) He believed that Tesla could get the same results with a cheaper suite of sensors. Musk first revealed that Tesla was considering adding autonomous-driving features in 2013. Tesla's ultimate strategy was to bypass lidar with a combination of simple cameras, radar (which uses radio waves to estimate distances to objects that are farther away), and ultrasound (which uses sound waves to estimate the distance to objects in the immediate vicinity). The cameras, which aren't much different from the ones used in smartphones, produce a video stream that is then analyzed by algorithms trained to recognize objects. Tesla's key partner for the camera system is Mobileye, an Israeli computer-vision company which claims that its software can detect vehicles up to two hundred and thirt y feet away using a single standard-resolution camera—all while adding only a thousand dollars to the price of a car.

In October of 2014, Tesla began offering its Model S and X customers a "technology package," which included this sensor array and cost about four thousand dollars. The equipment allowed the company to record drivers' movements, unless they opted out of the tracking, and—most important—to start amassing an enormous trove of data. A year later, it remotely activated its "Autopilot" software on tens of thousands of these cars. Suddenly, drivers had the ability to engage some limited autonomous functions, including dynamic cruise control (pegging your car's speed to the speed of the car in front of you), course alignment inside highway lanes, and on-command lane-changing. Some drivers were unnerved by the Autopilot functions, and cars occasionally swerved or drove off the road. But many of Tesla's tech-tolerant early adopters relished the new features.

Autopilot also gave Tesla access to tens of thousands of "expert trainers," as Musk called them. When these de-facto test drivers overrode the system, Tesla's sensors and learning algorithms took special note. The company has used its growing data set to continually improve the autonomous-driving experience for Tesla's entire fleet. By late 2015, Tesla was gathering about a million miles' worth of driving data every day.

To understand how commanding a lead this gives the company in the race for real-world autonomous-driving data, consider the comparably small number of lidar-based autonomous vehicles—all of them test cars—that some of its competitors have on the road. California, where much of the research on self-driving cars is taking place, requires companies to register their autonomous vehicles, so we know that currently Nissan has just four such cars on the road in the state, while Mercedes has five. Google has almost eighty registered in the state (though not all of them are in service); it is also doing limited testing in Arizona, Texas, and Washington. Ford announced earlier this year that it was adding twenty new cars to its test fleet, giving it thirty vehicles on the road in Arizona, California, and Michigan, which it says is the largest fleet of any traditional automaker. By comparison, Tesla has sold roughly thirty-five thousand cars in the U.S.  since October of 2014. The quality of the data that these vehicles are producing is unlikely to be as rich as the information the lidar cars are providing, but Tesla's vastly superior fleet size means that its autonomous cars can rack up as much driving experience every day or two as Google's cars have cumulatively.

With its approach, Tesla has been making a classic information-age wager: that software and processing can beat hardware. Lidar systems will no doubt continue to get cheaper: in February, Velodyne, the world's preëminent civilian lidar company, announced the release of a pared-down system costing about eight thousand dollars. But, even so, it will take time for suppliers to ramp up production to the point that carmakers can reliably buy lidar devices at a lower cost and deploy them widely. Software, by contrast, can be improved cheaply and continually, and can be updated remotely in an entire fleet of vehicles overnight. The genius of this system is that Tesla doesn't have to decide up front whether the Model 3 will be self-driving—it can just install the necessary hardware and make the cars autonomous at a later date.

This could prove particularly important given the profound regulatory and liability challenges that lie ahead for any company looking to create a fully autonomous car. The U.S. Department of Transportation classifies autonomous vehicles on a scale of zero (no autonomous functions) to four (no driver necessary). For both regulatory and technological reasons, Musk's fantasy of drivers fetching their autonomous cars from across the country by 2018 seems unlikely. Much less far-fetched is the idea that the Model 3 will be capable of full autonomy, but in limited settings—which would make it a level-three vehicle under the federal system.

Such a car would require some degree of monitoring by a driver, but under the right conditions—say, highway driving—it would allow people to mostly disengage from the task of driving in order to read, check e-mail, or watch a movie. (Some Model S drivers report being able to do so already, though from a safety standpoint they really shouldn't.) Fully autonomous highway driving would only be an interstitial step for city dwellers, but it would be very popular with suburban commuters—and it will be a major technological coup if Tesla beats Google, Nissan, and GM in the race to deliver. That said, from a commercial standpoint, the Model 3 is poised for massive success, regardless: more than three hundred and twenty-five thousand consumers have now placed deposits, amounting to fourteen billion dollars in potential sales. Autonomous or not, delivering that volume of vehicles on time, on budget, and on spec will be a challenge unto itself.

Sign up for the daily newsletter.Sign up for the daily newsletter: the best of The New Yorker every day.

Source: Will the Tesla Model 3 Be the First Truly Self-Driving Car?