Financial Times: Are Robotaxis a Losing Technology?

Recently, the esteemed British daily published a comprehensive analysis of the battle that is de facto taking place between traditional car manufacturers and technology companies for the future of autonomous vehicles, which is interesting from a Croatian perspective as it directly concerns robotaxis, questioning whether Amazon and Google have decided to support the wrong technology.

In simple terms, the tech giants have bet all their cards on the development of level 4 autonomous vehicles (level 5 is the highest level, complete ability for autonomous driving anywhere), while the auto industry has turned to so-called ADAS (advanced driver-assistance systems), which involves refining technologies that assist the driver (level 1 and 2 autonomy, or partial automation of vehicles), which is cheaper, implemented, already generating profit, and may also offer a path to higher levels of autonomy.

In simple terms, ADAS is currently focused on highway driving, while higher-level autonomous vehicles are envisioned as solutions for urban traffic, primarily in the form of robotaxis. The text cites arguments from tech companies that the gap between level 2 and level 4 is too great and insurmountable, but on the other hand, mostly small tech startups partnered with large vehicle manufacturers believe otherwise. As the author explains, until recently, the degree of technological development justified those betting on the development of highly autonomous vehicles, which includes robotaxis that the Croatian Project 3 Mobility intends to develop with the help of European funds, but this has changed rapidly due to the rapid advancement of ADAS technologies. Now, FT suggests, more than the belief in technological capabilities, the problem is basic economics. Level 4 vehicles are extremely expensive to produce and test, and their commercialization, or mass distribution in cities, primarily in the form of robotaxis, is even more costly.

Companies working on them are therefore ‘bleeding’ money and will continue to lose it in huge amounts for a long time because their commercialization will apparently not be so quick. On the other hand, those who have focused on ADAS, developing and improving vehicles to levels 1 and 2, are already profiting from it, and their cash flow prospects look far better. Moreover, completely opposite to the former.

It is, in fact, no longer a question of whether we will have highly autonomous vehicles, the text continues, but rather who will get to them first. A gradual approach within ADAS could ultimately be faster than a direct attempt to develop highly autonomous vehicles because the former has experienced a sudden boom, while the latter has faced unexpected problems. Google’s Waymo intended to order 82,000 such vehicles back in 2018, Uber thought it would have as many as 100,000 highly autonomous vehicles on the road by last year, and Lyft aimed to make most of its fleet autonomous by this year, FT writes, but none of that has happened. The closer they got to commercialization, the more complex the problems became.

Meanwhile, the tortoise has begun to catch up with the hare and ADAS now offers more and more advanced functions built into vehicles intended for customers, not test cars being studied by engineers. In other words, while these are still studying and researching, spending enormous amounts of money all the while, others are mass-producing, technologically advancing, and making profits in the process. To illustrate, FT provides the following data. Cruise has raised ten billion dollars so far and opened a new credit line of five to build advanced autonomous vehicles. Waymo has received large sums from Google for ten years, and last year it secured another 3.2 billion dollars in fresh capital, then an additional 2.5 billion last month. Zoox has practically run out of money, so it sold itself to Amazon at the beginning of the pandemic, and Apple has been working on its autonomous vehicle since 2014, but has yet to present even a prototype. Additionally, Uber has practically paid its rival Aurora to take over its autonomous vehicle team of 1,200 people, while Lyft sold its Level 5 department to Toyota in April.

ADAS, on the other hand, has become a gold mine, FT emphasizes. Revenues in that market amounted to 25 billion dollars and are expected to triple by 2027 according to estimates from BlueWeave Consulting. The situation is also better from a regulatory standpoint. For obvious reasons, regulators are not enthusiastic about granting permits for robotaxis, or vehicles with high levels of autonomy, but they have no problems with levels 1 and 2; on the contrary, they consider such vehicles a welcome increase in traffic safety and even require that new vehicles have some ADAS technologies built in.

Furthermore, even if ADAS takes a long time to reach higher levels of autonomy, that is not such a problem because for the reasons mentioned, they actually do not need to rush. They have revenue and the favor of regulators. However, for the other side, the story is significantly different. Given the costs, manufacturers of level 4 autonomous vehicles will soon have to start releasing larger quantities of these cars to generate some revenue. Additionally, further acceleration of ADAS vehicle development is likely because, with their growing number on the roads, the amount of real-time data used for further system improvements is also increasing. These data are currently, the text admits, inferior to those from level 4 vehicles, but that is likely to change because stable revenues allow ADAS vehicle manufacturers to afford increasingly better sensors, thus better data.

The only way for level 4 to overtake level 2, it seems, is rapid and mass commercialization, but analysts are skeptical. One tells FT that the shift from the suburbs where some of them currently operate to the entire country will not happen for another ten or 20 years. Another problem is the number of collisions that would begin to occur with the mass introduction of such vehicles. The text cites the head of Mobileye, a company owned by Intel, Amnon Shashua, who says that achieving an average of one collision every million miles, which is twice as good as a human, would be a reputational disaster at large numbers. ‘If I drive ten miles an hour, that means I collide every hundred thousand hours of driving. So if I distribute a hundred thousand cars on the streets, I will have a collision every hour,’ he explains the unpleasant math of the mass commercialization of level 4 vehicles.

If the ADAS approach overtakes the high-tech competition, it will be a very sweet victory for the auto industry, which until recently was under siege by technology companies that seemed poised to ‘take their bread’. Those who combine both approaches will fare best in all of this, but it seems that one of the two approaches will prevail, and despite the significant advantages of the ‘tortoise’ at this moment, the battle remains uncertain.