Tesla Autopilot Crash Knowledge Reveals Little To No Enchancment For 2023

Tesla lately launched a brand new replace to its Car Security Report, revealing knowledge for all 4 quarters of 2023.

The voluntarily printed quarterly experiences embody the variety of miles pushed per accident registered when utilizing Autopilot know-how, not utilizing Autopilot know-how, and the U.S. common (the latest knowledge accessible from NHTSA and FHWA for calendar yr 2022).

The info provides us glimpses of how Autopilot improves car security, however it has some limitations (extra on that on the backside of this publish) because of the methodology. For instance, Tesla counts all accidents, whatever the trigger, although greater than 35 p.c of all Autopilot crashes happen when the Tesla car is rear-ended by one other car, the corporate says.

In response to Tesla, the variety of miles pushed per accident registered in 2023, when utilizing Autopilot know-how, diversified from 5.18 to six.18 million miles pushed.

The outcomes are combined—21% worse than a yr in the past in Q1, 21% higher in Q2, 6% worse in Q3 and 11% higher in This fall. In different phrases, we will not actually say that there’s a clear, constant, and noticeable enchancment.

With out utilizing Autopilot know-how, Tesla automobiles statistically word extra accidents—from 1.00 to 1.52 million miles pushed per accident registered.

Not solely that, the outcomes have been worse than in 2022 in every of the 4 quarters of 2023 (Q1 down 9%, Q2 down 5%, Q3 down 11%, and This fall down 29% year-over-year).

There’s a important distinction between the 2 modes of driving (with or with out Autopilot) and the distinction continues to extend. Nevertheless, we should word that the 2 knowledge units shouldn’t be in contrast in opposition to one another, other than offering us with some concept of the numbers.

It is vital to notice that the outcomes are comparable solely inside a selected class (Autopilot or with out Autopilot), not between the classes because the enter knowledge is perhaps broadly completely different (like easy freeway driving or advanced metropolis driving). In different phrases, we are able to solely see whether or not the lively security techniques are bettering over time (and it is also solely a tough comparability), however we will not evaluate Autopilot to non-Autopilot driving.

We assume that the right use of Autopilot improves security, however Tesla’s report doesn’t permit us to judge the distinction.

Beneath we connected a full chart with all numbers supplied by the producer plus NHTSA and FHWA knowledge concerning the common distance between an car crash in the USA.

The distinction is fairly enormous, however as soon as once more, we should not evaluate the numbers immediately. The U.S. common features a a lot older car fleet (with fewer security techniques), which alone is an enormous issue. The U.S. common is 0.670 million miles (as of 2022 since there is no such thing as a data for 2023 but).

Tesla has its personal model of the chart:

Necessary elements:

• knowledge for every setting is perhaps collected at completely different driving situations (like easy freeway driving or advanced metropolis driving), which makes the outcomes incomparable between the classes
• outcomes is perhaps affected by numerous elements, together with seasonality (diminished daylight, climate situations), much less driving throughout lockdown
• NHTSA and FHWA common for the U.S. (up to date not often) consists of all vehicles, additionally outdated
• Tesla’s data concerning the methodology of registering accidents:
  “We acquire the quantity of miles traveled by every car with Autopilot lively or in handbook driving, primarily based on accessible knowledge we obtain from the fleet, and accomplish that with out figuring out particular automobiles to guard privateness. We additionally obtain a crash alert anytime a crash is reported to us from the fleet, which can embody knowledge about whether or not Autopilot was lively on the time of influence. To make sure our statistics are conservative, we depend any crash wherein Autopilot was deactivated inside 5 seconds earlier than influence, and we depend all crashes wherein the incident alert indicated an airbag or different lively restraint deployed. (Our crash statistics should not primarily based on pattern knowledge units or estimates.) In apply, this correlates to just about any crash at about 12 mph (20 kph) or above, relying on the crash forces generated. We don’t differentiate primarily based on the kind of crash or fault (For instance, greater than 35% of all Autopilot crashes happen when the Tesla car is rear-ended by one other car). On this means, we’re assured that the statistics we share unquestionably present the advantages of Autopilot.”
• assuming the methodology was not modified, we are able to see how every class improves over time