Mobileye CEO: Tesla accepting higher risk to lower its autonomy costs
The CEO of Intel-owned Mobileye, Amnon Shashua, gives us his perspective on Tesla, Waymo and the state of the self-driving industry.
Tim StevensFormer editor at large for CNET Cars
Tim Stevens got his start writing professionally while still in school in the mid '90s, and since then has covered topics ranging from business process management to video game development to automotive technology.
Mobileye CEO Amnon Shashua has an elevated perspective of the automotive industry, to say the least. As a technology partner and supplier to virtually every major auto manufacturer save
and Daimler, he is uniquely qualified to say who's doing well and who, perhaps, is lagging behind in the rapidly accelerating race towards true vehicular autonomy. He also has concerns that some players may be engaging in risky behavior.
It isn't just the major auto manufacturers that Mobileye, an Intel subsidiary since 2017, counts among its clients. The company's driver-assistance systems are desired by startups as well, and Shashua counts newbies such as NIO among his customers. There's also the world's most famous automotive startup,
, which is no longer a Mobileye customer, but once was.
"Companies like Tesla or NIO," Shashua told me, "they skip the use of a Tier 1. They become their own Tier 1. I think this is something that is possible to do when you have small scale... That allows them to move faster." Shashua was referring to the complex relationship between vehicle manufacturers and suppliers, a process with which many have become more familiar, thanks to Tesla's various struggles to get the components where it needs, when it needs.
Tier 1 suppliers are the most important, and they often furnish complex components like entire transmissions or even engines. Tier 2 are generally smaller and produce smaller, individual pieces to the Tier 1s. By skipping Tier 1 suppliers, Shashua says, automotive startups retain more control over the resulting product and can integrate more quickly.
"With Tesla," Shashua continued, "the first Autopilot was introduced in November 2014. It took about a year of development. With regular OEMs, it took closer to three years. That there are less players in the loop, in the chain, accelerates things."
Of course, moving quickly creates risk, and indeed, we've seen the ramifications of that with Tesla. It initially allowed its Autopilot system to function for extended periods even with no hands on the wheel, later dialing that capability back to seconds and disabling the Autosteer system entirely if a driver repeatedly ignored warnings.
Different companies accept different levels of risk, Shashua said, but given that many of his company's customers are on the more conservative side, Mobileye's systems must be well-nigh perfect. Half of the equation is defining how the car should behave in complex situations, a process detailed in Mobileye's Responsibility-Sensitive Safety, or RSS, which you can read more about in my impressions from my autonomous ride.
If you can guarantee the car will play by those rules, it then becomes a matter of how reliably the vehicle can see the world through its sensors. In engineering terms, it's garbage in, garbage out: You can't define a car's behavior if you can't rely on what it sees. "The way that you guard yourself against sensing mistakes," Shashua said, "is not only through algorithms that are well-tested, but also through redundancy. So now, the level of risk you are willing to take is comparable to the level of redundancy that you are willing to invest in."
Mobileye, though it initially developed its autonomy solution to use only cameras, is backing that up with a discrete lidar- and radar-based package, in theory providing full redundancy. Tesla, however, insists it will offer full autonomy without lidar sensors on its cars.
"So what is Tesla stating? They are willing to take more risks in sensing mistakes because they have less redundancy, because they want to have a lower-cost technology." This, Shashua said, contrasts with a traditional OEM. "Another car manufacturer can come in and say, 'I want to have almost zero risk in having a sensing mistake and I'm wiling to pay more for that, or my customers are willing to pay more for that, so therefore I'm putting not only radars and cameras, but also lidars.'"
As you can imagine, Shashua has thoughts about the current state of Autopilot, which Tesla rewrote internally after splitting with Mobileye in 2016. "You see that they had, I think still have, difficulties with reaching Autopilot 1.0. After three years of development, let's assume that they have reached parity now, it makes sense. This amount of time. This is where they have 10 times the computing power and eight cameras instead of one camera."
But don't read these comments as malice, merely pride in the decades of work that Mobileye has created. "I wish them the best of luck moving forward," he is quick to add.
Roadshow has reached out to Tesla for comment on Shashua's statements, and will update this story if we receive a reply.
And what about
, a company that is using parent company Intel's hardware, but not Mobileye's? "I really hope that they succeed," Shashua said. "I think in order to make a live industry, you need to have competition. So I hope they succeed, and I hope they succeed as early as possible."