So much happened in the auto industry this month it can’t fit in one column; I whittled it down to the things that interested me. There is a lot of variety by topic — from Auto Shanghai to a new operating system to an interesting DARPA story to the European Union proposal for AI use regulation. These ten stories are summarized in the following table.
Auto Shanghai is among the world’s largest auto shows and is the first auto show to be held after the pandemic. The attendance is expected to reach about 1 million people and around 1,000 exhibitors. The press days at Auto Shanghai were held April 19-20. There were lots of announcements and I have included a few.
As expected, there were many battery electric vehicle (BEV) introductions, with at least 20 new BEVs at the show, from Audi, BMW, BYD, Geely, Genesis, GM, Lynk, Mercedes-Benz, Nio, Polestar, Toyota, Volvo, VW and Xpeng. Toyota’s bZ4X’s BEV has an on-board solar charging system, allowing sunlight recharging when stationary.
There were two BEV flying cars or air taxis at Auto Shanghai. Geely showed an air taxi from its joint venture partner Volocopter. Xpeng showed its second prototype for a flying car called Heitech. The flying car has undergone eight years of development.
Baidu Apollo was prominent and unveiled upgrades to its intelligent driving and cloud solution. Baidu announced plans to have Apollo pre-installed on more mass-produced car models and bring Apollo Navigation Pilot (ANP) to more cities — 20 cities in 2021 and 100 cities by 2023. Apollo will accelerate the large-scale deployment of Baidu’s robotaxi business, Apollo GO. Apollo is also increasing development of solutions for commercialization of autonomous driving. Apollo is highlighting its milestone of L4 road testing exceeding 10 million kilometers.
ZF introduced a new version of ProAI, which is based on Nvidia’s Drive platform. The latest ProAI board is based on Drive Atlan with production availability in 2024. ZF has a family of ProAI boards that can use Xavier, Orin or Atlan. ZF has received several significant orders for its ProAI but did not specify its customers. ZF announced a production contract for its radar technology with SAIC that will start in 2022.
Toyota to use Apex.OS
Toyota’s Woven Planet is using Apex.OS, an easy-to-use software development kit (SDK) for automotive safety-critical applications, such as autonomous vehicle software. The Apex SDK can handle safety-critical applications and can speed up autonomous software development.
Toyota’s Arlene platform will use Apex. Arene is a Toyota software development platform with advanced tools, APIs and safety building-blocks to shortens software development time.
The origin of Apex.OS is the open-source ROS (Robot Operating System), which is popular for AV development. Apex has updated the ROS code to get functional safety certification. The rewrite project was successful and Apex.OS has been certified by TUV Nord for functional safety — ISO26262 ASIL D. It is rare that an open source software platform receives functional safety certification.
Toyota is an investor in Apex.AI along with Volvo Group, Hella, Airbus and several VC firms.
Walmart invests in Cruise
Walmart is investing an undisclosed amount in Cruise, which brings the size of Cruise’s latest funding round to $2.75 billion. The investment is part of Walmart’s effort to get a last-mile logistics network. Walmart and Cruise’s started last-mile test to deliver groceries in Scottsdale, AZ. In late 2020. The project is in an early stage with no expansion to other cities.
Walmart is also doing last mile delivery tests with Waymo, Ford and Nuro, which are continuing. Walmart has also used Uber, Lyft and Door Dash for goods deliveries.
Honda-AutoX partnership in China
Honda will collaborate with AutoX on AV research in China — including AV testing on public roads in multiple cities. Honda will adopt and leverage AutoX’s AV software platform and technology. Honda will use its Accord and Aspire vehicles for AV technology testing.
Arbe 4D radar design win
Arbe, an Israeli radar startup, was selected by AutoX to provide its 4D Imaging Radar Platform for AutoX L4 Robotaxis and other AV projects. Over the next five years, AutoX is planning to integrate 400,000 Arbe 4D radar systems in its Level 4 fleet with multiple radar units included for each L4 vehicles.
Arbe is in the process of a SPAC IPO that was announced in March 2021. The assigned value of Arbe in the SPAC agreement was around $720 million. Current Arbe VC funding is around $55 million.
Meituan Goods AV
Meituan, a leading Chinese shopping platform, is starting to operate a new goods delivery AV in Beijing. The new goods delivery AV is operating in L4 mode with loading capacity of 150kg with speed up to 20km/h. The previous version of the goods AV has completed a total of 35,000 deliveries.
According to Pandaily, Meituan is planning AV delivery services of food and vegetables in Beijing, Shenzhen and other locations. The Meituan goods AV is similar to Nuro’s R2.
Ocado invests in Oxbotica
Ocado is a UK online grocer that has been successful reselling its technology to other grocery chains to help it build and run its own online ordering and delivery operations. For instance, Kroger bought an Ocado warehouse automation system that recently became operational near Cincinnati in Ohio.
Ocado is investing nearly $14 million in Oxbotica to develop delivery AVs. This as a strategic Ocado investment to develop AVs that will work across its operations, within warehouses (likely first use-case), between warehouses or middle-mile deliveries and to last-mile grocery deliveries to people’s homes. Oxbotica is the leading AV software developer in the UK with over $90 million in VC funding.
Argo.ai AV safety report
Argo.ai released a new AV safety report in mid-April. It is quite extensive and discusses all of the 12 elements identified in the U.S. Department of Transportation’s Voluntary Safety Self-Assessment guidance. Argo is testing L4 AVs in six cities in the U.S. Argo probably has the most realistic perspectives on AV testing and safety. Here is a quote from the report:
“Fleet performance and safety are measured against three main metrics: Safety, Trip Quality, and Uptime. Within these broad categories, we monitor a number of key data points, including rates of potential collisions found through simulation and their estimated severity; critical failures of key SDS (self-driving system) features or systems; violations of road-traffic regulations; and quality and completion of ride. We also monitor the performance of our SDS, from software to sensor suite, to ensure that our automaker partners are able to deliver a viable mobility service”. The report is available at: ArgoSafetyReport.pdf
On another Argo subject, there are stories circulating that Argo is planning to go public.
DARPA SSITH project
DARPA SSITH (System Security Integrated Through Hardware and firmware) is a project on hardware security for developing secure processors. The goal is processors that are inherently immune to whole classes of hardware vulnerabilities that can be exploited by malware. In an April 2021 IEEE Spectrum article, one project was described that was very successful—a University of Michigan processor, called Morpheus. The IEEE Spectrum link is: Morpheus Turns a CPU Into a Rubik’s Cube to Defeat Hackers – IEEE Spectrum.
According to IEEE Spectrum, 580 cybersecurity researchers spent 13,000 hours trying to break into the Morpheus processor and all failed. If you need more information the above link will give you good perspectives. This is a few technical details of Morpheus from the IEEE Spectrum article.
Morpheus was based on a RISC-V processor architecture. Morpheus also uses the notion of undefined semantics, which means that any hacker is looking for information that is not known. Hence, Morpheus can detect an attack just by identifying what the hacker is looking at. Morpheus makes the underlying implementation of the machine—the undefined semantics—change every few hundred milliseconds. The key mechanism is making the machine’s undefined semantics change rapidly and never be the same again and it is done by simple and fast cryptography.
There were known vulnerabilities in the Morpheus code. Without Morpheus, it would have been trivial to get hacked. DARPA encouraged all participants to have known vulnerabilities, because the point of the program was to build hardware that could protect vulnerable software.
I don’t know if the Morpheus technology can be used in future processors for automotive applications, but I sure hope so—even if it is a long way to commercial products. I am convinced that cybersecurity will remain the toughest software and hardware issue for automotive electronics and we need to explore potential solutions.
European Union AI use regulation
The European Union proposed rules governing the use of artificial intelligence. Violations of the law could get fines of up to €30 million (over $36 million) or up to 6% of global sales — whichever is larger. The law is called the Artificial Intelligence Act and is available at this link: Proposal for a Regulation on a European approach for Artificial Intelligence | Shaping Europe’s digital future (europa.eu)
The impact of the new AI rules may be similar to the EU’s data privacy regulation, the General Data Protection Regulation. After the GDPR took effect in May 2018, it became the de facto privacy standard for most of the world’s largest companies.
The AI Act calls for a four-tier system that groups AI software into separate risk categories and applies an appropriate level of regulation to each.
- The top-level would be AI systems with an unacceptable risk to people’s rights and safety. Such AI systems would be banned. An example in this category is AI software that would allow governments or companies to implement social scoring systems similar to what China is using.
- The next category is called high-risk AIs. This category has a long list of included software and proposed limits. Examples that impacts the auto industry is road traffic and first responder services.
- The third category is for limited-risk AIs like chatbots. These programs disclose that you are talking to an AI so you can determine if you want to continue using the AI system.
- The fourth category is for AI systems with minimal risk to people. The vast majority of AI systems will be in this category. An example is spam filters. No regulation is planned for this category.
It will probably take around two years to get the proposed AI rules through the various steps in the EU government structures. It is likely there will be changes to the proposed regulation.