One day robo taxis will be safer than cars driven by humans, and when that day occurs, the world will change; that day is getting closer.
There are many who think autonomous cars will never happen because AI will never be able to out-compete humans. Such an argument overlooks a key point; humans are lousy drivers. For autonomous cars to be better than us wouldn't take much.
And the latest data on the testing of autonomous cars shows that the day when they are better than us is getting close. Meanwhile, in San Francisco, robo taxis are no longer a thing of the future; they are here now.
Yes, in the city known for its golden gate, Cruise — the autonomous cars subsidiary of GM — has been permitted to operate robo taxis between the hours of 10 pm and 6 am. They aren't allowed to drive over 30 miles per hour and can only run on certain streets. So yes, there are restrictions, robo taxis are not about to become omnipresent, but it is an ominous move for the traditional car industry.
Oh yes, and a man has to walk a few paces in front waving a red flag — no strike that was getting mixed up with the anti-car hysteria of my youth when horseless carriages arrived on the streets of London. "The trouble with these new-fangled motor cars," said Jacob Rees, my Dickensian brother, "is that they lack sentience — a horse would never voluntarily run into another horse, he explained to me in exasperation."
But Jacob, my boy, you were wrong — pretty sure I read that motor cars took off, and when I hear all those reasons why autonomous cars are a pipe dream, I think of you smoking your pipe and assuring me that the horse was the future and the workhouse was the only moral solution to poverty — or was that my sister Priti, I forget now.
Back to autonomous cars, Dr James Jeffs, all-round clever guy and analyst at IDTechEx, has been poring over the latest data on the testing of autonomous cars.
Last year he noticed that the disengagement rate on autonomous car testing —when the safety driver feels the need to take over — had been improving exponentially. Ever since 2016, the number of disengagements per 1,000 miles driven was halving every year. Dr Jeffs worked out that if this exponential rate continued, within a couple of years, the disengagement rate per thousand miles would be less than the number of crashes per thousand miles in a car driven by a human.
Last year, in a report, IDTechEX stated that by 2024 "autonomous vehicles will perform with the same safety as the average American", and by 2046 "autonomous vehicles will be able to cover the same distance as the entire US fleet (three trillion miles) between disengagements."
But that was last year; what does the latest data say?
Well, the exponential rate slowed. In 2020 and 2021, the number of miles per disengagement merely increased 1.5 fold.
Why was that? The problem was Waymo. As IDTechEX states: "Waymo have increased their testing mileage from 630,000 miles in 2020 to a staggering 2.3 million miles in 2021, quadrupling their previous effort." But as part of this increase, Waymo autonomous cars were tested more heavily in San Francisco.
It added that San Francisco "turned out to be a crucible of fire."
It "speculates the dense population, urban traffic, and abundance of cyclists and pedestrians have proven very difficult compared to the relatively green pastures that Waymo has been used to. This shows in their disengagement metrics which have fallen from 29,900 miles per disengagement to just 8,000 miles per disengagement."
IDTechEX says: "Despite this downturn in disengagement performance, Waymo is still making ground on the path to commercialisation. In Feb 2022, the California Public Utilities Commission permitted for Waymo to start charging passengers for rides in San Francisco."
However, "this service does still require a safety driver to sit behind the wheel and monitor the autonomous systems."
But what with the Cruise robo taxi plying the streets of San Fran too, Dr Jeffs reckons that "we can expect an Uber-like growth, but a slower pace." He said, "I think in the next five years, we'll see Cruise spread to tens of US cities and other service providers, like Waymo, come online."
Furthermore, comparing autonomous car disengagement with car crashes involving humans is "a tough comparison to make. Robotaxis have every incident logged; human drivers are only measured on official crash statistics, not all collisions are reported."
Dr Jeffs added:
"In 2021, California DMV recorded 116 collisions, of which 52 occurred in autonomous mode, and four could be attributed to the fault of the AI system. During this period, the testing fleet covered four million miles, corresponding to 1,012,000 miles per AI driver collision.
"This should be taken with a pinch of salt. There were several collisions that were put down to the safety driver at fault, where the safety driver took over immediately before the collision. It is likely that had the safety driver not been there, the AV would have caused a crash."
And the conclusion:
It may be too soon to draw a conclusion.
What we can say is that once it can be shown that robo taxis are safer than cars driven by humans, the global economy will be transformed. The cost of travel will plummet, but the traditional car industry, such an important part of the global economy, will be dealt a hammer blow.
Some say the day might never happen — that robo taxis will never be safer than human-driven cars, but then again, some said the motor car would never replace the horse-drawn cart.
The AI revolution is here
Jan 25, 2023
The impossible conclusion about technology becoming less disruptive and why it is so dangerous
Jan 20, 2023
Tech bubble! Are you kidding?
Jan 06, 2023