2021 Year in Review for Robocars


Here's my annual countdown of the biggest stories in self-driving for 2021. This year I have a summary of my text stories, and also made a video countdown because there's lots of nice background video to use.

You can read the text version on the Forbes site at:

Robocars 2021 year in review


I wonder if Tesla do find vision only, with strong input from the NHTSA, is not good enough. Maybe they will end up purchasing a LIDAR company to catch up and skip RADAR?

Tesla has charged hundreds of thousands of customer $10K to provide FSD. If they need a major hardware retrofit it's a loser. There is nowhere to easily put a LIDAR on Tesla cars with their glass roof, though there are some that go into the headlamps or in the rearview mirror but that makes it more difficult.

Fair call, although as I understand vision systems they can have a weakness in understanding the distance to stationary objects against the background. Even a basic forward looking LIDAR (or Radar) could give a last few seconds warning of a serious failure. I guess there are huge costs either in introducing it or not having it. Probably one for both Tesla the NHTSA to sort out.

No top-hats and skateboards for you.

Have you read David Twohig's opinion A Skateboard ain't a Platform.

The OEMs will defend their turf given the bruises from Tesla.

Most of his points are valid, though not all can't be solved. Certainly networking and power are done in more modern ways where they can be divided. But yes, the vehicle will need to be designed for strength as a whole, and the frame of the top hat won't just be quickly bolted on the base, I suspect the bases will be designed with a variety of structural bodies meant to be "on top of" (really structurally a part of) and the rest hung on that.

The skateboard companies want to be the big guys, who call the shots on their product and on what can be done with it. So you're not going to just buy it and screw something on top of it. If you're also big you can demand things you need in the skateboard, otherwise you take what you are given.

Or it might never work, as he says. That's the situation with Waymo and Geely. Waymo is doing overall high level design, Geely making the full structure of the vehicle, but leaving places on it for Waymo to bolt on sensors. I suspect Geely will run the wiring to Waymo's spec, or just leave conduits for them to do that. Waymo will want to do only what they feel they must do, which is sensors, networking, internal screen. They will be happy to leave the rest to Geely. Waymo is a big enough customer that many aspects of the vehicle will be Waymo's design though. And I suspect Waymo will own the VIN, not that it matters because nobody will ever buy the vehicle.

responses always appreciated.

The VIN implication is a good addition to the analysis. This is for liability reasons mostly?

A modern car is made by many different supplies. Big car OEMs may make almost none of their car, but they design it and sell it, using contract manufacturers. But they "own the vin" which makes them responsible for liabilities on it. Or sometimes they do the final assembly, or rarely they make everything important in it.

But in a vehicle that will never be sold, just used as a taxi, that does not matter as much.

The new website for the Yole Group, including Yole Intelligence and Yole SystemPlus goes live.

Yole Group - 18x as many L1-L2 Light Vehicles In Operation versus L4-L5 Light Vehicles by 2040.

See pdf -
Imaging for Automotive 2022
Technology and Marketing report

Especially slides 11 and 12 - nuances in slides, must extrapolate (expand slides for scaling)
NOTE: slides are cumulative but unit scaling different yet useful

Interesting trends or growth rate of all L2 types versus L3L4L5 for 20 years (extrapolate distribution levels, rate of change, absolute uptakes and comparative percentages over select time periods)

Average yearly uptake of various ADAS levels in absolute values, cumulative and comparative percentages over select time periods.

Attrition rate of non-ADAS (eg a minimum of 270 million non-ADAS light vehicles to be replaced by ADAS LVs between 2020 and 2030).

Eg With somewhat linear growth rates, average of 67 million L1-L2 ADAS LVs estimated to be manufactured from 2020-2027.

Eg. The delta of the estimated 4.7% VIO penetration rate for L4-L5 by 2040 (100 mil) with the 89% VIO penetration rate for L1-L2 by 2040 (1.8 bil) is the essence of the incrementalism versus moon-shot debate and what is the most accessible and quickest path to safety for all.

Where Are The Autonomous Cars?
on semiengineering dot com

Is TOPS estimate clickbait?

ZEEKR First Customer for Nvidia Drive Thor, With Initial Production Vehicles Planned for Early 2025.

Geely-owned automaker ZEEKR announced it will integrate DRIVE Thor on its centralized vehicle computer for its next-generation intelligent electric vehicles, starting production in early 2025.

Intel's loss of business for Zeekr's next gen AV had to be known in advance.

NVIDIA Unveils DRIVE Thor, includes automated and assisted driving, parking, driver and occupant monitoring, digital instrument cluster, in-vehicle infotainment (IVI) and rear-seat entertainment into a single architecture.

ZEEKR has already selected DRIVE Orin for its in-house advanced driver-assistance development. With the forward-compatible DRIVE platform, ZEEKR can seamlessly leverage its current development investments on DRIVE Orin to DRIVE Thor for its long-term production roadmap and beyond.

Nvidia is scrapping the Drive Atlan system on chip ahead of schedule for Thor,

Is the lack of self certification in Europe part of the impetus for Thor?

Thor also seems like an attempt by Nvidia to mitigate the lack of L2+ / Premium ADAS platform design wins of late.

Does the trend of in-house chip design and software IP by OEMs present a challenge to Nvidia silicon and/or platform adoption?

Is Nvidia attempting to thwart off Chinese acceleration in silicon design for premium ADAS or AVs (eg Baidu or Horizon)?.

Jim McGregor from IntelTechTour on Twitter

2 Mobileye EyeQ 5 AV demo rides - an 11-camera system in a Ford Focus and a radar/lidar/camera robotaxi. Both worked, but the Robotaxi was smoother with higher object recognition and tracking.

Reading the above comment, one assumes the Focus was SuperVision and not Drive.

Zeekr- Intel, then Waymo, now Nvidia - the Chinese manipulation never ceases

add PonyAI as company partners with Geely affiliates to build robotaxi fleet

Baidu & Geely invest EV joint venture Jidu Auto

The ebb and flow of winning next generation L4 design wins.

July 2022
"In the meantime, our business is thriving, we are growing on all fronts and the future was never brighter" CEO MBLY on IPO update

September 2022
Zeekr selects Nvidia for L4 for 2025 production.

Zeekr & MBLY only started L4 AV project exploratory 9mos ago 1/2022.

Will MBLY and Zeekr now even bring to market a consumer L4 car in 2024?

Reminiscent of NIO flip flop? BMW?

And to think Zeekr obtained a US$100 million investment from Intel Capital only a year ago (2021)

Geely musical IP: from Intel/MBLY, Baidu/Jidu, Waymo, PonyAI, Nvidia, ..

All eyes on Ford BlueCruise with MBLU SuperVision tech first.

"ZEEKR will .... as well as developing new lidar-based features with Mobileye."

The intentional wording in this statement is impossible to overlook.

The notion of the addition of Lidar in a L2 + Premium stack?

from September 2020
This long-term agreement will extend to enabling ADAS features on numerous models across multiple brands under the vast worldwide umbrella of Geely

www dot mobileye dot com/blog/our-new-deal-with-geely-is-a-game-changer-says-shashua/

from September 2022
Yesterday's Mobileye and Geely Holding Group are expanding their collaboration to bring Mobileye SuperVision to three additional Geely brands, with additional plans to bring the technology to other models in the future.

basically, SV technology has been approved for production in Zeekr 001 in a forthcoming OTA update and then rolls out further next year to other brands. Nothing about L4 despite the lidar reference.

today feels like Qualcomm is challenging Waymo through proxies.

Is a delineation between AVs and robotaxis becoming more pronounced?

Add new comment