CEO Amnon Shashua embarks on a premature AV victory lap in advance of Mobileye's IPO.
At last we have a winner. Autonomous driving is solved, and the champion is Mobileye,claims company CEO Amnon Shashua. Robotaxis are coming this year, and “Level 4 consumer AVs” in 2024. Easy-peasy.
So, are Argo, Aurora, Cruise and Waymo chopped liver? Or are Mobileye’s promises simply too good to be true?
Let’s go “under the hood” with Amnon Shashua during Mobileye’s CES 2022 keynote, beginning with Shashua’s CES slide deck. Mobileye’s owner, Intel, which intends to take Mobileye public this year, also provided a summary of CES announcements, of which there were many. A video of Mobileye’s CES presentation is here.
To accurately report Shashua’s presentation, a transcript was obtained from YouTube. Punctuation was added, sentences formed, technical errors corrected. (We sought comment from Intel. It has yet to respond.)
Zeekr, robotaxis, consumer AVs
The “Level 4 consumer AV” is a future model to be launched by Zeekr (pronounced “Zeek-er”), a little-known brand from Chinese automaker Geely. Zeekr is described as “a new electric mobility technology and solutions brand satisfying the global demand for premium electric vehicles.”
Thus, a wannabe Tesla competitor.
At 15:22 in the CES video, Shashua defines Mobileye’s offering:
“There is a new category emerging in driving assist, it’s kind of a premium ADAS. We call it L2+. It’s based on full-surround sensing, like in the Zeekr we have eleven cameras, seven long-range cameras (8 megapixel) and four parking cameras. There are cloud-based enhancements, using our [Road Experience Management, REM] maps. There is a firmware over-the-air update involved, both for upgrading functionalities, updating functionalities and also streaming the map data to the car, providing a cloud-based enhancement.”
Shashua clarifies: “This is Level 2, you know there should be eyes-on and mind-on by the driver.” He continues:
“The second emerging pillar is the Level 4 robotaxi. It’s geofenced. The cost of the system could be in the tens of thousands of dollars, and this is our robotaxi platform that we unveiled back in September..
“A new category that’s emerging in the 2024-2025 timeframe. This is a Level 4 consumer AV. What it has, that the robotaxi does not have, one is that it should drive everywhere, and this is where our mapping technology, the REM mapping technology, becomes very, very, critical. So, it’s not just geofenced capabilities…and the cost, the MSRP [Manufacturer’s Suggested Retail Price] to the customer around $10,000, means that the cost, a bill of materials cost, should be way below $5,000 to support a consumer AV.”
Thus defined, a “Level 4 consumer AV” can drive everywhere (is not geofenced) and has a bill of materials cost “way below” $5,000.
At 17:41, Shashua adds:
“When we look at autonomous driving, you know, the future can move in two directions. One is the robotaxi, moving people, moving goods. Mobileye is active there, and many, many, companies are active, like Waymo, Argo, Cruise, Aurora, etc.
“Then there’s another possibility, which is a consumer AV. You purchase a car, the car is enabled, has a Level 4 capability. You press a button, and you can have your eyes-off and mind-off the driving experience. Mobileye is active there, [so too is] Tesla of course.”
This is a perfect example of Autonowashing. It is not possible today to purchase any vehicle offering “eyes-off and mind-off driving.” It may be possible sometime in the future, but sometime is a timeline no one can predict.
Shashua then asserts that the consumer AV and robotaxi futures are not equivalent. At 18:38, he observes:
“That means the consumer AV future contains inside it also the robotaxi, because if you can purchase a car, then of course you can add to it the layers of ride hailing, and sell it to public transport operators, and transportation network companies and provide the service.”
Entering full IPO sales mode, Shashua throws shade at Mobileye’s robotaxi competitors:
“The other way is not clear. That means, if you have a robotaxi technology which is geofenced and expensive, this does not lead to consumer AV. For consumer AV you need geographic scalability, and you need self-driving system cost way below $5,000. We argue that for a consumer AV, this needs to be purpose-built. You cannot start with a robotaxi technology in terms of scalability, lack of scalability and high cost, and bring it to a consumer AV. It needs to be purpose-built, and this is what Mobileye is doing.”
Which begs the question: Why does Mobileye continue to develop robotaxis at all? This issue isn’t addressed.
Capabilities, robustness, efficiency
At 20:31, Shashua pivots:
“Let’s look at the criteria for a good solution that can cover both futures. We divided them into three pillars. One is capabilities, you would like to support a very wide self-driving operational design domain, whether it’s highway, rural, urban, arterial roads, to provide a very, very, wide [operational design domain]; and human-like driving policy, it needs to merge in traffic in a very smooth manner, similar to the way a capable human driver would merge into traffic.”
“The next pillar is robustness. It should have a very, very high mean time between failure. Much, much, higher than the average human driver. Somewhere 10-times higher, 100-times higher, 1,000-times higher, but it should be higher. And third is efficiency. Efficiency is cost, as I said we want the compute and sensors to be way below $5,000, and also support scale being able to drive everywhere.”
The presentation then morphs into a Vegas magic show worthy of David Copperfield or Penn and Teller. At 21:36, Shashua says:
“In terms of capability, we support full operational design domain, full ODD, from Level 2 to Level 4. That means we claim that the differentiating factor between a Level 2 to Level 4 is not the ODD, it is the MTBF. It is: What is the mean time between failure?”
Pay close attention, because here comes the trick:
“If the mean time between failure is lower than the average human driver statistics, then it’s a Level 2 system. If it is higher, then it can be a Level 4 system.”
With that, let’s unpick the trick.
Shashua doesn’t mention SAE International anywhere in the keynote, nor does he make any specific references to SAE J3016 (the levels of driving automation). There’s no definition provided for “failure,” nor of “average human driver.” Which average? Average where? Average when? Shashua is unilaterally redefining terms, and the new definitions are vague.
Then, something else.
On Jan. 4, Intel issued a Mobileye-Zeekr press release, stating:
“Mobileye and Zeekr today announced plans to expand their strategic technology partnership with the goal of developing a new, all-electric consumer vehicle with Level 4 (L4) capabilities.”
Mobileye’s headline-grabbing claim of a “consumer L4 AV in 2024” had been diluted to “the goal of developing … Level 4” before Shashua’s pre-recorded comments had even been transmitted. Mobileye’s problem is that everyone else has the same goal of developing Level 4 capabilities, so this wasn’t news.
All of this reasonably equates to the marketing messages of a company preparing for an IPO, but none of it to a company making tangible progress towards its objectives. Circumstances are exacerbated by the fact that these claims have been made before by Shashua and Mobileye, in a BMW press release dated July 1, 2016:
“BMW Group, Intel and Mobileye are joining forces to make self-driving vehicles and future mobility concepts become a reality. The three leaders from the automotive, technology and computer vision and machine learning industries are collaborating to bring solutions for highly [SAE Level 4] and fully [SAE Level 5] automated driving into series production by 2021.”
How did that collaboration work out? In November, BMW went with Qualcomm. This helps explain why Intel now seeks a Mobileye IPO.
For what purpose is J3016?
Shashua’s keynote succeeded in highlighting a critical issue: “Level 4” is the new “IoT.” Investors love it. Great headline. But no one quite knows what it actually means.
Let’s turn to Phil Koopman, associate professor at Carnegie Mellon University, to help us out.
Koopman details the division of responsibilities between the driver and automation for each of the levels. The summary slide from the tutorial is shown below, which highlights the fundamental problem with J3016: It is a terminology-standards document, not a safety standard.
Yes, the AV industry really fooled lawmakers and safety agencies for years with promises of saving lives and radically improving road safety by using a standard defining terminology–not safety. Koopman summarizes the key issues in a video, and rightly identifies regulation and trust as the key challenges ahead.
In a blog post, Koopman proposes replacing J3016 with four categories of vehicle operation: Driver Assistance, Supervised Automation, Autonomous Operation and Vehicle Testing.
He notes there is intentionally no mapping to the SAE Levels, “because that would import baggage that could compromise safety.”
We now wait to see which AV developer is first to publicly endorse and adopt Koopman’s proposals.
Safety-leader Argo AI is clearly the front-runner.
In a recent interview, Koopman explained the unpalatable truth facing the AV sector: “There’s nothing I’ve seen showing whether AVs will be safer than humans in the short to medium term. Machine learning is brittle, and it struggles with things it hasn’t seen before.”
For many years the AV industry has demanded little or no regulatory oversight of its activities, translated to: “Let us do whatever we like.” Increasingly, that attitude has morphed into: “Let us say whatever we like.”
AV technology innovators continue to push the boundaries of safety on public roads with neither consequence nor rebuke. They must remember there is peril and potential legal liability in over-inflating claims.
At CES, Mobileye’s Shashua took a Level-4 victory lap for a race no one has yet won, having deftly switched rules midway to suit Mobileye. In so doing, he exposed the levels of driving automation as little more than a marketing tool.
SAE International needs to withdraw J3016 and replace it with a standard that is robust, unambiguous and safety-oriented.
Koopman’s proposal looks like the perfect start.
This article was originally published on EE Times.
Colin Barnden is principal analyst at Semicast Research and has over 25 years of experience as an industry analyst. He is considered a world expert on market trends for automotive vision-based driver monitoring systems (DMS). He holds a B.Eng. (Hons) in Electrical & Electronic Engineering from Aston University in England and has covered the automotive electronics market since 1999.