Autonomous Driving — The Current State of Play
The autonomous vehicle (AV) landscape has evolved significantly since I wrote my previous article on the AV State of Play in March 2017. AVs have passed the peak of the hype cycle: some disillusion and bad press, i.e. the recent Uber and Tesla accidents. There have also been gaps in the communication between technical and business leaders as to when Level 4 (L4) and 5 will actually be widely accessible. Will it be 10 or 15 years before vehicles and infrastructure can provide a level of safety that surpasses the 1 death per 100M km human drivers achieve today on any road? No one can say. Nevertheless, soon the deployment of AVs will begin for specific car and truck use cases and will continue progressively. Huge efforts are being undertaken to enable this revolution, including the emergence of a whole AV development ecosystem. Not surprisingly, legislation seems to progress more slowly than technology. Let’s look at the latest state of play on these key dimensions.
Race to develop or acquire AV tech
Two megatrends absorb a very significant portion of automotive incumbents’ R&D budgets, whether OEMs or Tier 1: electrification and autonomous driving. New players, i.e. tech giants and startups, concentrate clearly on the latter. Since the deployment of AV will be tightly linked to shared mobility (see below), the race and the collaboration are intense between the “old guard” and new players.
OEMs continued building in-house AV expertise which will likely be used mainly for integration purposes. They have also established multiple partnerships to assemble full solutions (software and software) and test them. Ford and GM have made significant acquisitions, respectively with ArgoAI ($1bn over 5 years) and Cruise ($600m), to jump start their in-house development. Toyota, which said it would not give into autonomous driving, changed their mind and are playing catch-up. Other OEMs are progressing on their chosen paths. As for Tier 1s, Delphi and Autoliv have carved out their new tech business (mainly AV-related) and Continental is considering the same move. Aptiv (ex Delphi) bought nuTonomy for $450m a few months ago. Stakes are high!
Among ride sharing companies, Uber is the most advanced though they hit a few road blocks during the past year: a legal battle vs Waymo, a deadly accident and the resulting removal of a test license. Conversely, Lyft created its own AV development team, Level 5, which partners with multiple OEMs and Aptiv. Building on its AV expertise, Waymo is moving into autonomous mobility on demand (AMoD). The Alphabet company is also working with Honda to develop vehicles fit for purpose and with JLR to integrate its AV tech and operate the vehicles. Waymo likely has the most mature solution, leading the pack in terms of system disengagements with 0.1 per 1,000 km in 2017, i.e. the backup driver must take over on average once every 10,000 km.
In China, Didi has accelerated its AV efforts in the last months. The world’s largest ride hailing company has been testing AVs on public roads in China. Didi also recently created the “Auto Alliance” platform to bring OEMs in its ecosystems and co-design vehicles fit for AMoD (See my Apr 2018 article for more on this subject). Earlier this year, Didi opened a lab in Silicon Valley to work on AV tech and was granted a permit last month to test AVs in California.
It is clear that the AV race is on between China and the rest of the world. The Middle Empire now prohibits foreign companies from mapping out the country, which ultimately creates a barrier to entry. Chinese tech giants are also developing AV tech. Baidu created the Apollo “open source” AV development platform and intends to supply not only the software but also the hardware as seen at CES. Tencent and Alibaba are also investing in this space.
On the sensor front, Lidars continue to attract much attention as they are required for L3 and above, except if you listen to Tesla. Whereas spinning models (e.g. Velodyne, Valeo, Ouster) see their projected price drop, a number of companies are working on theoretically cheaper solid state versions, targeting a 3-digit — $ or € — price at scale. We have yet to see any volume though. Valeo’s “Scala” is the first one to go into production for Audi A8’s “Traffic Jam Pilot”. Recently, a production contract was announced between BMW, Innoviz (solid state) and Magna, which I suspect has been brought in to hold the startup’s hand — the Tier 1 invested in Innoviz in 2017.
An ecosystem has emerged around AV tech development
AV tech testing and validation is a very painstaking task, given the quasi infinite number of corner cases and the safety objective stated earlier. Road testing is time consuming, expensive and does not necessarily present all the required test conditions. Simulators allow to quickly test fleets on a large number of adjustable scenarii, making it the de facto complement to road testing.
Last year, Waymo was reportedly “driving” 12M km a day on their proprietary simulator, Carcraft, i.e. even more than the 10M km they have driven on the road since 2009. In parallel, simulator startups have emerged, e.g. Metamoto or AIMotive. Others, such as DeepenAI or MightyAI, provide the labeled, real life data (i.e. people and objects are identified on all images) that feed simulators. Further up the value chain comes the collection of real life data, most economically done through crowdsourcing. While focusing on safety, Nauto’s cameras installed on fleet vehicles are well positioned to provide such data. Likewise, Tesla has been collecting similar input for many years, which provides them with a huge amount of data with which to test their AV stack.
The legislative framework progresses heterogeneously
Governments now realize that AVs are definitely coming. They also increasingly understand that adapting the legislative framework can bolster innovation and economic development. Whereas the U.S. has hosted a large part of the AV development to-date, the country as a whole does not present a unified framework. Last fall, the SELF DRIVE and AV START Acts were introduced in the U.S. Congress to create a structure. In parallel, individual states have enacted their own legislation. Four states, i.e. Michigan, Arizona, California and Ohio (plus Pittsburgh), now allow AV testing on public roads. 53 companies have obtained a license to test AVs (with backup drivers) in California, vs 27 as of March 2017. Since April 2018, companies can also request permits to operate AVs without backup drivers. So far, two companies have applied, including Waymo.
In Europe, France already allows permit-based testing with a backup driver and plans to enable full driverless vehicle testing on public roads in 2019. The UK adapted its legislation last fall: developers can apply to test their vehicles on public roads without a human operator. Germany passed a law last March to allow AV testing with a backup driver, while clearly assigning liabilities.
In China, several cities have passed local laws to enable permit-based AV testing. Realizing the strategic importance of AV development in the global race, the central government just issued national guidelines for testing self-driving cars as part of the “Made in China 2025” plan. Korea has been quietly enabling its own AV development, allowing testing on a defined set of public roads, with the objective to have AVs for sale by 2020. In Singapore, where the first ever self-driving taxis picked up select passengers in 2016, AV tests on public roads started in 2015.
The deployment will be progressive
Last year, OEMs were announcing the introduction of L4 AVs by 2020-2021 — and in 2019 for Tesla. As of mid-2018, the vehicle with the highest level of automation is the Audi A8. Its L3 feature, “Traffic Jam Pilot,”, provides conditional autonomy up to 60 km/h. However, it will only be sold in Europe due to the heterogeneity of the global legislative framework.
As far L4/L5 deployment, the earliest commercial applications will be AMoD operating in geofenced areas where HD 3D maps have been developed and the AV system has been trained. Waymo will reportedly begin operating a self-driving ride-hailing service in Phoenix later this year — the company has been transporting non-paying passengers for a few months there. Uber, Didi and Lyft are poised to test similar services though they are running behind Waymo.
Concurrently, Navya and EasyMile had an early start with automated shuttles and continue to test their complete solution, transporting “real passengers” in geofenced areas around the globe. May Mobility emerged last year to do the same thing, along with Local Motors. Zoox, with $290m raised in two rounds, came out of stealth mode about a year ago with the ambitious plan to design, manufacture, maintain and operate fully autonomous vehicles, competing with both OEMs and ride-sharing companies. Operations should start in a large US city (SF?) in 2020.
As far as OEMs are concerned, last January GM made the biggest promise so far, presenting a version of its Chevrolet Bolt EV sans steering wheel and pedals! They intend to use it in a commercial ride-share service at scale starting in 2019. To this end, GM petitioned the U.S. Dept. of Transportation for permission to begin operating such a vehicle by then.
Trucks also go autonomous
Autonomous driving is also making its way into the trucking world. Truck makers have been working on this for some time, e.g. market leader Daimler. More recently, Uber (with the acquisition of Otto in 2016) and Waymo have been investing in applying the AV tech to specific use cases. Several startups are also making significant progress, namely TuSimple, Embark, Starsky Robotics or to a lesser extend Peloton Technology (focus so far on truck platooning). The low hanging fruits are port terminal tractors (close environment, controlled traffic) and long stretches of freeway from entrance to exit (single direction, no intersections). This will allow drivers to focus on short hauls, thus avoiding long absences from home.
It’s still a long journey ahead. In the meantime, hands on the wheel and eyes on the road!
Managing Director, Orsay Consulting
Feel free to like or comment this article on LinkedIn (https://goo.gl/HyWwp7)