Machine Learning and Data Are Fueling a New Kind of Car
The automobile is being dismantled, reimagined, and rebuilt in Silicon Valley.
Intel’s proposed $15.3 billion acquisition of Mobileye, an Israeli company that supplies carmakers with a computer-vision technology and advanced driver assistance systems, offers a chance to measure the scale of this rebuild. In particular, it shows how valuable on-the-road data is likely to be in the evolution of automated driving.
While the price tag might seem steep, especially with so many players in automated driving today, Mobileye has some key technological strengths and strategic advantages. It’s also developing new technologies that could help solidify this position.
Mobileye uses a single camera, together with a proprietary computer chip and some clever software, to provide various advanced driver assistance features. Its systems can, for example, identify the speed limit from road signs, or identify vehicles and pedestrians for an automatic braking system.
David Keith, a professor at MIT’s Sloan School of Management who studies technology adoption in the automotive industry, says besides offering a simple, low-cost solution, Mobileye has amassed a huge amount of data—something that is vital to the machine learning that underpins automated driving today. “Their technologies are highly reliable, honed over millions of miles of driving experience, which competitors cannot easily replicate,” he says.
It isn’t hard to see why Intel should want to enter the auto market. The increasingly capable computers, sensors, and wireless connections now found in vehicles are enabling big changes across that industry. Meanwhile, Intel has seen its position of dominance eroded in recent years as desktop and laptop computers fade in importance, and as different types of computer chips have become more popular. Competitor Nvidia has already captured a sizable share of the growing auto market.
Keith adds that Intel will aim to use its hardware expertise to develop the increasingly sophisticated fusion systems—combining cameras, radar, and possibly laser sensing, or lidar—needed bring fully automated vehicles to market.
If your car is capable of identifying a road sign or a pedestrian on the road ahead, there’s a good chance it already uses one of Mobileye’s chips for the task. The company’s vision systems are a simple, low-cost solution that offers surprisingly sophisticated sensing.
The company therefore offers Intel a good way into the automated driving market, which promises to grow as the technology matures in the coming years.
For its vision system, Mobileye employs deep learning, a machine-learning technique that has given computers powerful new capabilities in recent years. This involves capturing images as cars drive around, and annotating them to identify things like road markings, traffic signs, other vehicles, and pedestrians. The images are fed into a big neural network, which is tweaked until it can reliably recognize the relevant elements of an image. If Mobileye’s system is unable to identify something, it’s usually possible to simply annotate some new images and add them to the learning data set.
This isn’t to say that it’s perfect, or all that’s needed for automated driving. Tesla had been using Mobileye’s vision technology for its Autopilot semi-automated driving system until last year. The companies ceased working together after a fatal accident involving a car controlled by Autopilot. In the fallout from the crash, the carmaker criticized the vision system provided by Mobileye. Executives from Mobileye countered that its technology was never meant to be used in this way.
Technology now under development at Mobileye could help automated cars drive more safely in the future. In December, I met with Amnon Shashua, Mobileye’s CTO, and Shai Shalev-Shwartz, VP for technology. They explained how Mobileye is now using reinforcement learning, a technique inspired by the way animals learn through experience, to teach computers how to drive safely in complex and subtle situations (see “10 Breakthrough Technologies 2017: Reinforcement Learning”).
As part of this effort, Mobileye is developing a simulated driving environment to enable learning. It hopes this will become the standard environment for testing automated driving software. They also explained that Mobileye is working with several carmakers on a way for them to share the data collected with other companies for a price. This could help accelerate (no pun intended) progress toward fully automated driving.
The result could be a transformation of transportation as we know it. Indeed, the prospect of profound disruption has caused a stampede for technology and talent among automakers, suppliers, and startups.
Stephen Zoepf, executive director of the Center for Automotive Research at Stanford, agrees that Intel’s acquisition of Mobileye shows how critical data and machine learning are to the auto industry’s future. But he adds, “It’s also evidence of the degree to which demand for talent is outstripping supply in the autonomous vehicle space.”
Geoffrey Hinton tells us why he’s now scared of the tech he helped build
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
ChatGPT is going to change education, not destroy it
The narrative around cheating students doesn’t tell the whole story. Meet the teachers who think generative AI could actually make learning better.
Meet the people who use Notion to plan their whole lives
The workplace tool’s appeal extends far beyond organizing work projects. Many users find it’s just as useful for managing their free time.
Learning to code isn’t enough
Historically, learn-to-code efforts have provided opportunities for the few, but new efforts are aiming to be inclusive.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.