In my previous column I used a pictorial view of the competition and cooperation among the developers and future users of autonomous vehicle software driver platforms. I also gave information and perspectives on five key suppliers: Waymo, Cruise, Argo, Aurora and Mobileye.
In this column, the second part of my AV Software Platforms analysis, I am discussing a few more AV software suppliers. I use the same figure as before, but this time with some additional relationships called out.
The figure of the AV software driver platforms is complex; by necessity the chart you see below is a simplification, as there are many additional companies participating. The cooperation relationships are primarily drawn to the main auto OEMs. A few relationships that have been cancelled are still marked with a dotted line and a red X.
Nvidia is primarily known for providing the GPU chips for AV computer hardware and is the clear leader currently. But Nvidia is also supplying a lot of AV software with a focus on the software tools and platforms that are needed to develop, test and verify the AV software platform that its customers develop. The key to Nvidia’s success is the CUDA platform. CUDA is a parallel computing platform and application programming interface model. CUDA provides a software layer that gives direct access to the GPU’s virtual instruction set and parallel computational elements.
The CUDA platform became the leading platform for developing AI algorithms and applications using machine learning and neural networks. Nvidia’s GPUs could handle the large and growing computing performance that AI applications needed via many generations of compatible GPU processors. Similar dynamics are also propelling the AV hardware and software industry.
Nvidia also realized that AV software development and especially the training of AI models is difficult to do on PCs and need to be developed on large IT systems or cloud systems such as AWS, Azure, Google. Nvidia is supplying IT systems that us GPUs that are compatible with the GPUs that are used in Nvidia based AV systems in cars. This makes it advantageous to develop and test AV systems using Nvidia GPUs. Nvidia is especially strong in testing for virtual driving miles by combining its AV hardware and software platforms that its customers can use.
At its recent virtual GTC conference, Nvidia described its web-based MagLev 2.0 cloud platform for AV software development. The GTC presentation shows how different the AV software development is compared to traditional automotive software development such as an infotainment or ADAS application. The key difference is the amount of sensor data—in petabytes—that has to be learned and built into the AI-based algorithms for the AV inferencing software that AVs must use.
Nvidia has many Tier 1 customers that have announced L3 and/or L4 AV hardware platforms including Aptiv, Bosch, Continental, Veoneer, ZF and many others. Nvidia has lots of OEMs that are using or planning to use Nvidia AV processors including Mercedes-Benz, Toyota, Volvo, Volvo Trucks and many others. Several AV software platforms are also based on Nvidia hardware including Aurora, AutoX, Baidu Apollo, TuSimple, WeRide and others.
Nvidia has and will continue to develop AV software for robotaxis and other AV use-cases. Nvidia is not planning to compete in the AV software platform market segment with Waymo, Cruise, Argo and Aurora and others for robo-taxis and other AV use-cases. Instead the AV software development companies can leverage Nvidia’s AV software platforms and software development tools for their own use.
Amazon has been busy investing in autonomous vehicles for several years — not a surprise considering Amazon is one of the largest logistics companies in the world. Amazon is probably shipping around 10 billion packages globally this year. Amazon invested in Aurora in February 2019 as part of a $530 million VC investment round. The Aurora Driver is planned for goods AVs, autonomous trucks and robotaxis and Amazon could use the first two right away. Amazon has also developed its own sidewalk AV for last mile deliveries. Amazon Scout started package delivery field tests in January 2019 in Washington. Amazon was a partner in Toyota’s e-Palette concept project. Amazon is also testing with autonomous truck startup Embark.
The biggest Amazon AV investment is the acquisition of Zoox for $1.2 billion in June 2020. Zoox has a different approach for its robo-taxi by developing both an AV software platform and a custom vehicle architecture with 4-wheel steering and no distinction between going backward and forward. Covid impact has lowered future VC investments in AVs and Zoox started looking for an acquisition partner. Amazon looks like a good parent for Zoox. Amazon is planning for Zoox to continue developing its robotaxi software platform. Amazon is also expected to use the Zoox software platform for goods AVs. Amazon can also leverage the specialized vehicle architecture for future goods AVs.
Amazon will have some future AV options. Amazon can use Zoox or Aurora AV technology or both. Maybe Aurora for autonomous trucks and Zoox for small goods AVs.
Motional is the name of Aptiv and Hyundai’s joint venture that is focused on robotaxi use-cases. Aptiv and Hyundai’s combined contributions total $4B, with each owning 50% of Motional. Motional plan to develop an AV software driving platform for robotaxi providers, fleet operators and automotive manufacturers for deployment in 2022.
Motional operates over 100 AVs in multiple cities including Las Vegas, Boston and Singapore. In January 2020, the AVs had completed over 100K robotaxi trips with safety drivers present. The Las Vegas robotaxi trips were done with Lyft as a partner.
Motional is primarily based on Aptiv’s AV technology. Aptiv was early in acquiring AV startups such as Ottomatika in 2013 and nuTonomy in 2017. Aptiv did an excellent job of leveraging these acquisitions into a strong AV software platform for robotaxi use-cases.
Four ride-hailing companies are included in the above figure — Lyft, Uber, Didi and Yandex. They have different AV strategies and operate in different regions with some overlaps.
Uber has focused on development its own AV software platform since 2015. Uber has spent around $2.5 billion on its AV investments according to a downbeat story by The Information.
With Covid straining Uber’s resources due to big drops in its ride-hailing business, developing its own AV software platform is a risky strategy. It would probably be better to use the Waymo Driver or another software platform. Note that Uber has ownership stakes in both Didi and Yandex from the previous sale of Uber’s ride-hailing business in China and Russia, respectively.
Uber acquired Postmates for $2.65 billion in July 2020 to greatly increase its future delivery business. Deliveries have increased dramatically during Covid and is likely to retain many of these new customers. These deliveries are currently via many types of vehicles—from bicycles to motorcycles and cars. Goods delivery market is projected to be a major AV segment in a few years. Focusing on goods delivery AVs would be a better strategy for Uber.
Lyft is the second largest ride-hailing company in the U.S. Lyft AV development launched in July 2017 and started testing in California in November 2018. Lyft had 19 AVs testing in California in 2019, based on California DMV data and drove 43,000 miles in 2019. Lyft is testing AVs via simulation and this became a focus during the Covid shutdowns in the Spring of 2020. No Lyft data on virtual testing miles are available.
Lyft’s AV strategy includes cooperation with AV software companies. Lyft has close cooperation with Motional, the Aptiv-Hyundai joint venture and has completed 100,000+ robotaxi trips with safety drivers in Las Vegas. GM invested $500 million in Lyft in January 2016 with a goal of future robotaxi cooperation. The investment may lead to future AV collaboration and use of the Cruise AV software platform.
Lyft is also focusing on the delivery opportunity and started its Essential Deliveries in April 2020 to help out its drivers as the ride-hailing business dropped after Covid hit. Lyft further expanded its delivery business through cooperation with GrubHub starting in October 2020.
Yandex is the leading internet search company in Russian and Yandex also is the leading ride-hailing operator in Russia and many surrounding countries. Yandex started AV development in 2017 and is now creating an AV subsidiary called Yandex SDG (Self-Driving Group). Yandex will invest $150 million in Yandex SDG and have 73% ownership. Uber will a passive ownership of Yandex SDG at 19%.
Yandex SDG has over 130 AVs and considerable AV road testing at over 4 million miles. Yandex is primarily testing in Russia but also in Israel. Yandex started to test in the U.S. in Ann Arbor, Michigan in July 2020. Yandex SDG plans to start robotaxi pilot service in Ann Arbor and Tel Aviv within the next year. Yandex SDG also has a sidewalk delivery robot called Rover which is in pilot testing in Russia.
Didi Chuxing is the world’s leading ride-hailing company with 550M users and 10 billion + trips per year. Didi started its AV research in 2016 and has over 200 people in AV technology R&D. The Didi AV group became an independent company in August 2019. In May 2020, Didi received VC funding of $500+ million for its AV subsidiary including $400 million from SoftBank. Didi has announced a most aggressive AV goal and said it plans to have 1 million robotaxis on its ride-hailing platform in 2030.
Chinese AV Platforms
China has many important AV software platform companies. Since this was described in a recent column: China Robotaxi: Full Steam Ahead, I will only do a short summary. The main Chinese AV software platform companies are Baidu Apollo, AutoX, Pony.ai, WeRide and Momenta.
Baidu Apollo is the most important AV software platform in China. It is an open source platform that has large support from Chinese companies and foreign companies that participate in the Chinese automotive industry.
In July 2020, Baidu Apollo surpassed 6 million kilometers (3.73 million miles) in urban road testing. Baidu’s fleet of 500+ AVs has tested in 27 cities worldwide including cities in California. Baidu Apollo has 150 AV licenses for AV testing in China including 120 permits that allow the AVs to carry passengers. Baidu has completed over 100,000 passenger-carrying trips as of June 2020. Baidu calls its robotaxi service Apollo Go and the service is free for now.
AutoX has AV activities in the U.S. and China. AutoX launched a goods AV service delivering groceries in a limited area of San Jose, California in 2018. In 2019 AutoX had one of the best results in California DMV’s AV testing results. In California, AutoX received the third AV license for operation without safety drivers.
In China, AutoX has robotaxi services operations in several cities including Shanghai, where it has 100 AVs on the road. AutoX is cooperating with several taxi operators and multiple auto OEMs such as BYD, Dongfeng and FCA.
At CES, 2020 AutoX announced a partnership with FCA to developing robotaxis using Chrysler Pacifica and AutoX’s AI Driver software platform.
Pony.ai is active in AV development in the U.S. and China. It has an impressive VC investment at over $760 million. In February 2020, Pony.ai received $400 million in investment from Toyota. Pony.ai and Toyota teamed up on AV pilot testing in August 2019 in China. Pony.ai also have partnerships with Hyundai and GAC.
The Pony.ai AV fleet of 100+ vehicles have traveled more than 2.5 million kilometers (1.6 million miles) in China and the U.S. combined. Pony.ai is testing in multiple cities in China— Beijing, Guangzhou and Shanghai.
WeRide is operating AVs primarily in Guangzhou where it started robotaxi testing in November 2018. WeRide has accumulated 2.8 million kilometers of AV testing and has served 90,000 robotaxi passengers. WeRide has 120 AVs with 40 AVs used in robotaxi operations and the remaining AVs used for testing.
WeRide received a permit in July 2020 to launch China’s first AV tests in Guangzhou without a safety driver but with teleoperation as a backup.
Momenta was founded in 2016 and is focusing on developing AVs using cameras and HD mapping technology, but without expensive lidars—at least for now. This is a strategy similar to what Tesla is doing. Momenta said its solution can generate HD maps with 10 centimeter-level relative accuracy. Momenta collected 2 billion kilometers of driving data for HD maps in 2019. Momenta’s technologies will be deployed and updated using crowdsourcing of data from fleet vehicles enabled by Momenta. This is similar to Mobileye’s REM strategy.
Based on its mapping platform data, Momenta is developing L3 and L4 AV software driving platforms that can be used on a variety of AV hardware platforms. Momenta received an AV road test license in Suzhou, China and has started AV testing. Momenta is planning robotaxi trial operation by the end of 2020.
Momenta has received VC investments of over $200 million. This include investments from Daimler in 2017 and Toyota in 2020. Majority of the VC funding is from Chinese companies with Tencent being the most interesting company due to its influence in internet related industries.
Other AV Platforms
There are many other AV software development companies shown in the above figure, but I will only comment on Tesla and Apple.
We don’t hear much about Apple and most of the information about Apple is speculation. This is my interpretation of the speculative information. Apple was initially planning to develop its own BEV-based autonomous car. This (too) ambitious planned was scaled back to develop an AV software driver platform plus a hardware platform including sensors and computer hardware.
Apple acquired Drive.ai in June 2019 as the company announced it was shutting down. Drive.ai started a free robotaxi service using fixed pickup and drop-off locations with seven AVs in Frisco, Texas in July 2018. Drive.ai was using limited AV teleoperation during its six-month trial operation. It looks like Apple acquired valuable technology and people expertise for a good price in this acquisition.
Where Apple is going next is hard to predict. My guess is that it is using a strategy similar to Aurora and/or Waymo. Apple could enter the robotaxi business sometime in the future. Apple could possibly buy another AV startup if or when VC investment for AVs slow down. Apple certainly has the financials to acquire a leading AV software platform such as Aurora if available. Maybe Apple should have made a bid on Zoox earlier this year, since the Zoox business model fit Apple’s need to control most of the activities in its businesses.
Tesla is the opposite of Apple—we hear too much and some of the information is too optimistic and not likely to happen in the specified time frame. Tesla has been adamant that it will not need to use lidars for its L4 autonomous vehicles—cameras and radar will be good enough. The rational is that if the human eye can do this for a human driver, the Tesla AV system using cameras and radar and AI software can do the same sometime in the future. The vast majority of experts in the AV industry believe lidars are required for L4 AV deployment in the near future. Some AV experts believe that major AI technology breakthroughs are needed before Tesla’s goals are possible and the timing of that is uncertain. I think the best case for Tesla’s AutoPilot is to move from L2+ to L3 where the car will drive itself in select areas, but the driver must be ready to take over on short notice. Tesla need better driver monitoring system for L3. Moving to L4 will require significant AI technology advances.
More “Egil’s Eye” from Egil Juliussen:
- AV Software Platforms: Cooperation & Competition
- AV Software Driver vs. Human Driver
- China Robotaxi: Full Steam Ahead
- Autonomous Vehicles in China
- Complex Automotive Software: What’s Your Strategy?
- Automotive Software Platforms: Current Status