Predicting consumer AV in 2025, Mobileye's CEO outlined strategies to make AVs “scalable, transferrable, economically viable and legislatively acceptable."...
Mobileye boldly predicted the launch of consumer AV in 2025. Speaking at CES, Amnon Shashua, CEO of Mobileye (which is owned by Intel), outlined strategies intended to make testing, technology and business development for autonomous vehicles “scalable, transferrable [to different geographical regions], economically viable and legislatively, socially acceptable” for years to come.
As car OEMs and AV startups today face the hard reality of long-term “zero revenue” from the Level 4 and 5 AV business, many have already shoved their AV projects onto the back burner.
But not Mobileye.
Shashua, speaking at Monday’s CES briefing, articulated measures the company has initiated, ranging from driving policy to HD mapping and redundancy.
These steps are markedly different from rivals’ approaches. Shashua said Mobileye is determined to “keep the oxygen flowing” while allowing to stay in — and win — the AV marathon.
During the media presentation, Shashua actually guaranteed a consumer AV launch in 2025. The introduction of commercial robotaxis will take place prior to that. Doing robotaxis before consumer AV introduction is critical, in his view, because it provides the opportunity to “practice” and work with regulators on AV rules of the road.
Shashua called robotaxi “a game changer” that will significantly reduce transportation costs. However, he added that with the launch of consumer AVs in 2025, changes will be radical. “All hell will break loose.”
Three differentiated strategies
“There are not many, but a few things we did differently,” said Shashua. All these deviations were part of Mobileye’s effort to solve the hardest problems with AVs up front.
First was the launch of REM — Road Experience Management Technology. Introduced five years ago, REM automatically creates maps that reflect “road changes in a timely manner, and provide centimeter-scale accuracy for cars driving on the road,” according to Mobileye.
Most AV companies build HD maps using designated mapping vehicles running around in every city. In contrast, Mobileye uses “existing cars equipped with EyeQ4 driver assistance systems,” according to the company.
REM, said Shashua, isn’t an event recorder. He explained that conventional HD mapping, by recording all the objects on the road through brute-force data collection, requires too much time and effort. “Such an approach wouldn’t scale. Mapping itself becomes the bottleneck.”
Instead, Mobileye’s REM limits recording to the “semantics of roads,” such as drivable paths, four-way stops and unprotected turns, he noted. This makes the carmakers’ collaboration with Mobileye easier, with carmakers collecting limited data (10 kilobytes per kilometer) for upload to the cloud. Currently six carmakers are working with Mobileye, by using the existing technology, covering 8 million kilometers per day. This has resulted in Mobileye having mapped almost 1 billion kilometers globally.
Mobileye subscribes to the idea that AV driving must be demonstrably superior to human driving. Shashua offered a quick back-of-the-envelope calculation:
The annual number of miles driven in the US is about 3.2 trillion. The number of accidents with injuries is about 6 million. Assuming an average speed of 10 mph, the mean time between failures [MTBF] is 50,000 hours of driving.
Consider deploying 100,000 AVs to serve as robo-shuttles. Assume each robotaxi drives five hours a day. So, with a 10x MTBF design, there should be one accident every day. With a100x design, an accident every week, and at 1,000x, one crash per quarter.
Getting every car on the road to score a 10x better MTBF would be a huge achievement. But for a fleet operator, an accident occurring every day is unacceptable.
Hence, Mobileye believes a lower boundary of 1,000x on MTBF “is a must.” To get there, collecting billions of hours of test driving is untenable.
Mobileye is advocating “internal redundancy” on a single sensor modality, instead of promoting “low-level redundancy” between different sensory modalities.
A year ago at CES, Mobileye disclosed its unique approach to internal redundancy by running different algorithms on camera-only systems. This year, Mobileye is unveiling plans for a separate subsystem consisting of a front-facing lidar plus a cocoon of radars performing the same level of safety accomplished by the camera-only system. By fusing the two subsystems — each independently developed and validated — a robotaxi should be able to achieve “true redundancy,” claimed Mobileye.
Mobileye’s camera-only perception system has been designed into Level-2 ADAS cars by Geely in China, with volume production starting in 2022, creating a new revenue stream before the fully autonomous vehicles start making money.
In the first phase of a robocar implementation, Mobileye is adding Luminar’s lidar and a cocoon of radar chips to its camera-only perception system.
For phase 2 sensor fusion in robocars, which starts in 2025, Mobileye will be ready with a home-grown lidar, by using frequency modulated continuous wave (FMCW) technology. Shashua called FMCW “a technological leap” that uses “Doppler instead of time-of-flight.” Leveraging Intel’s silicon photonics expertise (fab, manufacturing, and IPs), Mobileye expects in 2025 a FMCW-based lidar system-on-chip (SoC) designed by Intel for use in AVs.
Mobileye is also developing imaging radars that will be “software-defined, added Shashua.
Third, the most important technology Mobileye has developed thus far for AVs is RSS – responsibility-sensitive safety, according to Mobileye’s CEO. “This is our crown jewel.”
Using a set of mathematical formulas and logical rules, RSS formalizes human notions of safe driving, such as driving cautiously or driving recklessly. Fully autonomous vehicles sharing the road with human drivers must understand what people regard as careful. An AV can’t anticipate the unexpected, but it can be programmed to assume the worst-case scenario.
RSS becomes the foundation for the safety argument about robocars when AV suppliers deal with regulatory bodies. “RSS needs to be transparent,” said Shashua. “Safety can’t be a secret agent.”