Page 8 - EE Times Europe Magazine | June2020

8 EE|Times EUROPE

Full AV Stacks: Five Years In, Does the Progress Measure Up?

ment of Motor Vehicles (Figure 2).
By law, anyone actively testing self-driving
cars on California roads must disclose both
the number of miles driven and the number
of miles driven between “disengagements” —
those fraught moments when a human driver
is forced to take the wheel. The DMV formally
defines disengagement as “deactivation of
the autonomous mode when a failure of the
autonomous technology is detected or when
the safe operation of the vehicle requires that
the autonomous-vehicle test driver disengage
the autonomous mode and take immediate
manual control of the vehicle.”
Some safety experts believe that counting
the miles between disengagements tends to
encourage test operators to minimize their
interventions, skewing the results and even
putting the test drivers at risk. “Disengage-
Figure 2: Evaluating AVs by counting disengagements can encourage bad behavior on the ment is the wrong metric for safe testing,”
part of test drivers, some safety experts warn. (Image: IHS Markit) said Phil Koopman, a Carnegie Mellon
professor and co-founder of Edge Case
Research.
previously announced a plan to expand ware platforms. And of course, there’s Tesla, Nevertheless, AV disengagement reports
its R&D program with Hyundai to develop building its own full AV stack. help the industry to peer into the readiness
a self-driving platform. Apparently torn The notables in this crowded bin are level of autonomous vehicles, if not determine
between two lovers, Hyundai has said its new Waymo, Aurora, Argo AI, Intel/Mobil- their safety per se, argued Juliussen.
joint venture with Aptiv-nuTonomy won’t eye, Nvidia, and Drive.ai. Last June, Apple As of the end of last year, 65 companies had
affect its affair with Aurora. But it has been snatched up Drive.ai — a Mountain View, test-driving permits from the California DMV.
reluctant to kiss and tell, divulging few details California-based startup founded in 2015. Juliussen noted that while 567 vehicles were
about what it is actually doing with Aurora. Baidu’s Apollo is an open-source AV plat- “qualified,” only 420 AVs were on the streets.
form with a large ecosystem of developers. A chart provided by Juliussen offers a plot
OEM PLATFORM line showing changes in the annual disen-
In the bin for OEM platforms, Juliussen listed SCANT PUBLIC DATA gagement numbers of each AV platform from
GM-Cruise, Hyundai, VW, Ford-Argo, BMW, The toughest question of all in the AV market 2015 to 2019. But he questioned the validity
Mercedes-Benz–Bosch, Volvo, and Toyota. All is whose AV software platform is solid, ready of Baidu’s disengagement data. Baidu last
have been testing their own autonomous vehi- to go, and better than others. Given the lack year reportedly drove 108,300 miles and told
cles but vary in where they stand on AV stacks. of disclosure from the developers, we turned the California DMV that its vehicles went
Car OEMs with their own AV software plat- to another source of data that provides insight 18,000 miles between disengagements. Many
forms include GM-Cruise (built on Cruise’s AV into the AV landscape: the AV disengagement industry watchers, including Juliussen, are
software platform), Ford-Argo (based on Argo reports produced by the California Depart- skeptical because Baidu appears to be saying
AI’s full AV stack), and Toyota (developed
in-house). Volvo might have its own stack, but
it announced previously a partnership with
AImotive, a Budapest, Hungary-based soft-
ware platform supplier (formerly AdasWorks).
BMW is married to Intel/Mobileye’s AV
software platform. Volkswagen, without
its own AV software platform, last summer
jilted Aurora and turned to Ford’s Argo.ai. As
mentioned earlier, Hyundai has an apparently
bigamous, possibly precarious bond with
Aurora and Aptiv-nuTonomy.
HIGH-TECH SOFTWARE PLATFORMS
In the bin that includes the software-driven
high-tech segment, Juliussen lists Waymo,
Aurora, Argo AI, AImotive, and Drive.ai in
the United States and an Asian contingent
that covers Preferred Network in Japan, Baidu
(with the Apollo project), AutoX, Momenta,
WeRide, and Pony.ai.
Leading AV chip suppliers Nvidia and
Mobileye are also developing their own soft- Figure 3: Companies are still targeting L4 readiness by 2022. (Image: IHS Markit)

JUNE 2020 | www.eetimes.eu