Autonomous vehicles could be not so autonomous

As I explained in my last entry, autonomous vehicles had an intense summer, especially Waymo and Cruise, owned by giants Google and General Motors, respectively. Two of the most advanced startups in the race to dominate this technology.

On August 11th, 2023, both companies obtained approval from the state of California to start offering autonomous cab services 24/7 and charging (until then, their activity was restricted to a few hours of low traffic density). The city of San Francisco was close at hand.

However, their joy was short-lived. A few days later, on August 19th, the state authorities penalized Cruise after some serious failures the night before. First, a robotaxi crashed into a fire truck responding to an emergency. In another incident a few hours later, it was a "regular" car, piloted by a human, that ran a traffic light and ran over one of these robotaxis. In total, Cruise had to reduce the number of vehicles he had driving around the city from about 300 to half that number.

However, the General Motors startup was still missing the final straw, and it came a few days ago, when the state of California cancelled its autonomous vehicle license after one of its models failed to detect that it had a person run over under it. In the aftermath, Cruise announced it was canceling all of its U.S. operations. And Waymo has opted to slow its pace of expansion.

The harsh reality is that the startup that does not circulate is the startup that does not generate profits. A business and technology race to burn money, show results, and continue burning investors' money until they run out of patience or one of the startups emerges as the winner.

Where are we now, then?

Hundreds of autonomous vehicles have been driving around cities for hundreds, and that’s a huge milestone. Waymo and Cruise have driven a combined total of nearly 13 million km without a driver, including more than 6,4 million km in San Francisco since the beginning of 2023. Since California law requires autonomous vehicle companies to report every major accident and make it public, we know a lot about their performance.

In total, the two companies have reported just over 100 accidents involving driverless vehicles. It may seem like a lot, but they occurred in about 10 million km of driving. That's equivalent to one accident every 100,000 km, about five years of driving for a regular motorist. However, these statistics do not include other important incidents, such as ones involved with dogs, car hits with other vehicles where the autonomous vehicle flees, or sudden spontaneous combustion.

If we look back, what BigTechs promised us was cheaper, faster and safer city transport. One of the most famous critics of this technology is researcher Filip Piekniewski, who has already warned that after 100 billion dollars invested in this technology, 6 companies have gone bankrupt, 3 are trying to survive and only two seem to be doing well.

Piekniewski's attitude is probably too radical. In my case, I am convinced that autonomous driving will be achieved, although we may need more radical changes.

In many media you can find the data that between 90 and 94% of road accidents correspond to human error. The history of this figure dates back to no less than 1979. Subsequently, the US road safety authority, the NHTSA, published a two-page document in 2015 stating that the critical reason, which is the last event in the crash causal chain, was assigned to the driver in 94% of the crashes”. Nevertheless, the agency highlighted in the same text that, although the critical reason is an important part of the description of events leading up to the crash, it is not intended to be interpreted as the cause of the crash nor as the assignment of the fault to the driver.

The argument of replacing the driver with an algorithm to reduce that 94% of human causes is, therefore, false. And numerous experts in road safety have raised their voices about this deceptive use of statistics.

Therefore, maybe autonomous vehicles should not reflect the number of accidents as their main metric, but should have their own metrics. And it's probably also important to remember that every “autonomous vehicle” on public roads has a human safety driver on board or somewhere observing and controlling remotely.

A few days ago, the New York Times published an article that makes Cruise and these safety drivers blush

Autonomous vehicles might not be so autonomous

According to NYT journalists, the autonomous vehicle company employed an average of 1.5 safety drivers per vehicle, and they had to intervene in driving very frequently. If this were true, it would be logical to question whether the rest of the robotaxis startups are doing the same "cheats”. In that case, we would still be very far from the autonomous driving derived from popular imagination.

Perhaps, autonomous driving is not the solution to our traffic problems, but we must think about technologically assisted driving, which complements humans and helps them not to fail. Perhaps the technical feasibility of vehicles so complex their achievement is impossible in the urban mobility frameworks of our cities. It is probably also important to understand that robotaxis are not just a technological invention, but very much a socio-economic one.