We finally have some data on ADAS!
For too many years, the National Highway Traffic Safety Administration (NHTSA) took a laissez-faire approach to advanced driver-assistance systems (ADAS) and automated-driving systems (ADS). That changed in June 2021 when the U.S. road safety regulator issued an order requiring automakers, suppliers, and ADS developers to provide data on crashes where these systems were active before or during a crash on public roads. NHTSA recently released a trove of data collected in the first eleven months of this effort.
The purpose of the mandate was to build an understanding of the efficacy of automation technology ranging from SAE Level 2 systems that control speed and steering to fully automated vehicles. Longer-term, the agency may build on this data-gathering effort, with the aim to (hopefully) begin a rule making process that leads to setting some standards for this technology.
ADAS began appearing on mainstream vehicles nearly a decade ago and despite the proliferation of basic versions of the technology, our roads have become less safe. It’s clear that without standards, these technologies have been more of a feature checklist item than an actual safety benefit.
A dive into the actual data reveals some fascinating insights. The data was split into two groups, one for vehicles with L2 ADAS and the other for L3 to L5 ADS. To the surprise of no one except the hardcore Tesla fans who believe Elon Musk’s claims about the brand’s AutoPilot and “Full Self-Driving,” Tesla vehicles accounted for 70% of the 392 reported ADAS crashes. A genuine surprise, however, was seeing Honda reporting 90 crashes. It’s debatable if Honda even should have been included, since the Honda Sensing ADAS package doesn’t tightly integrate lane-keeping assistance and adaptive cruise control under a common control regime, as other automakers are doing. However, the company submitted data, so it’s worth a look.
Unfortunately, I believe NHTSA made a mistake in allowing companies the option to flag the crash descriptions as “confidential business information (CBI).” Anything flagged as CBI was redacted from the published dataset. Tesla continued its history of being opaque when discussing statistics for its AutoPilot driver-assist system, so all 273 records had the crash descriptions redacted.
Honda, conversely, left the crash descriptions involving its vehicles open to read. Given that there has been no media reporting of issues with Honda ADAS, I thought it was worth scrutinizing. It turns out that although LKAS or ACC was active just before or during the impact, it does not appear to have been a factor in any of the crashes. Instead, most were incidents of being struck by another vehicle at an intersection, being rear-ended or something else irrelevant to ADAS operation.
Since GM has never previously provided any data on crashes involving its hands-free Super Cruise, it was interesting to see two reports included in the NHTSA data. In both cases, the Super Cruise system was manually overridden by the driver seconds before impact. In one case, a driver slammed on the brakes before striking a deer. In the other, the driver admitted hitting the steering wheel with a knee that caused the system to deactivate. The vehicle then ran into a median.
There are opportunities for both the industry and regulators to learn from these reports. Understanding the scenarios involved could lead to improvements in the technology. But going forward, NHTSA should no longer allow companies to redact crash description information – this data is too important to our understanding of the technology and manufacturers should not be allowed to keep the scenarios private.
We also need telemetry data on when ADAS activates to prevent a crash. Hopefully, this will lead to some common-sense regulation of ADAS and ADS sooner rather than later.