On Wednesday morning, the National Highway Traffic Safety Administration released data on the safety, or lack thereof, of advanced driver assistance systems. Since June 2021, automakers have been required to inform NHTSA if any of their vehicles crash while using partially automated driving systems, also known as SAE level 2 systems.
As many suspected, Tesla’s Autopilot system accounted for the majority of crashes since the reporting period began. In fact, Teslas represented three-quarters of all ADAS crashes—273 out of 367 crashes reported between July 2021 and May 15, 2022. The news provides yet more data undermining Tesla’s safety claims about its Autopilot system.
In the past, Tesla and even NHTSA have claimed that Autopilot reduced crash rates by 40 percent. However, as we reported in 2018, that claim fell apart once a consulting company called Quality Control Systems got its hands on the data.
Autopilot’s woes did not end there, however, and a string of Tesla Autopilot crashes eventually spurred NHTSA’s Office of Defects Investigation into action. In August 2021, NHTSA’s ODI opened a probe into 11 crashes in which Teslas using Autopilot crashed into first responder vehicles and upgraded that probe to a much more comprehensive Engineering Analysis earlier this June. This is in addition to a separate NHTSA investigation into the propensity of newer Teslas—which lack forward-looking radar—to spontaneously brake. The investigation began in February.
Tesla’s Autopilot system has also been repeatedly singled out by the National Transportation Safety Board, which even blasted NHTSA for failing “to recognize the importance of ensuring that acceptable safeguards are in place so the vehicles do not operate outside of their operational design domains and beyond the capabilities of their system designs.”
Under NHTSA’s new standing order, automakers (and operators, if appropriate) have to file incident reports on any crashes involving level 2 ADAS that occurred on a publicly accessible road in the United States or its territories.
In fact, they must report any crashes if a level 2 ADAS system was operating within 30 seconds of the crash and the crash involved “a vulnerable road user2 or resulted in a fatality, a vehicle tow-away, an airbag deployment, or any individual being transported to a hospital for medical treatment.”
(This may be meant to address the concern that Tesla has programmed Autopilot to shut down if it determines a crash is imminent. Last week, NHTSA said that in an investigation of 16 Tesla crashes, several showed that Autopilot gave back control of the car to the human driver “less than a second before the crash.”)
The data in NHTSA’s latest dump is subject to a few caveats. Access to crash data varies by manufacturer, and some incident data may be incomplete, the agency says. Some crashes may be reported more than once if the data came in via the manufacturer but also as a complaint. The agency also notes that some data has been partially redacted to protect personally identifiable information or confidential business information and that the summary incident report data is not normalized; it’s merely represented as an absolute number of crashes.
The most common source of crash data came from telematics; complaints and claims were the second most common source. Of the 402 crashes broken down by reporting entity, Tesla was far ahead, with 273 crashes in less than a year. The second-greatest number of crashes involved Honda’s vehicles—in February, NHTSA opened an investigation into the Japanese automaker’s ADAS after receiving nearly 300 complaints about phantom braking. Subaru was third, with just 10 crashes over the same time frame.
Listing image by Tesla