Facebook Twitter

First self-driving crash data released, but questions of safety remain

What the first large-scale study on self-driving car crashes means for manufacturers and consumers

SHARE First self-driving crash data released, but questions of safety remain

In this Feb. 15, 2017, file photo, pedestrians and cars move along a busy section of North Avenue near the Georgia Tech campus in Atlanta.

John Bazemore, Associated Press

The National Highway Traffic Safety Administration on Wednesday released crash data that has been collected after a standing general order was issued in June 2021. The order enacted reporting requirements on manufacturers of automated driving technology, in order to “ensure that their vehicles and equipment are free of defects that pose unreasonable risks to motor vehicle safety.”

Though the information in the report gives a glimpse into autonomous vehicle use in the country, much more data will be required to determine the risks of these systems.

ADS vs ADAS systems

Autonomous features in a vehicle are classified on the Society of Automotive Engineers (SAE) scale from 0 to 5. On the low end are “momentary driver assistance” features, such as lane departure warnings. Level 2 systems can perform steering and braking while the driver is fully engaged. Levels 3-5 are not available for consumers yet; the system performs driving tasks with variable human input.

Automated driving systems (ADS) are systems classified as SAE Level 3 and greater. Manufacturers and operators of ADS vehicles were required to report accidents where ADS was in use within 30 seconds of the crash, and property damage or injury occurred.

Advanced driver assistance systems (ADAS) that are classified as SAE Level 2 were only required to report the crash if certain conditions occurred: fatality, injury, air bag deployment and others.

What the reports say

Both reports collected data between July 2021 and May 15, 2022. Each crash may have multiple reports submitted.

ADAS report (Level 2)

  • Three hundred sixty-seven crashes, 392 reports.
  • Average of 34.8 crashes per month.
  • Most commonly damaged on the front.

Data quality on the crashes is poor. 294 reports do not record injury severity. Of the 98 that do, there were 6 fatalities, 5 “serious” injuries, and 22 “moderate” injuries.

The most common reporting entities were:

  • Tesla (273 crashes).
  • Honda (90 crashes).
  • Subaru (10 crashes).

ADS report (Level 3-5)

  • One hundred thirty crashes, 258 reports.
  • Average of 12.9 crashes per month.
  • Eleven crashes involved a “vulnerable road user” (seven cyclists, two motorcycles, two electric scooters).
  • Crashes resulted in one “serious” injury, three “moderate.”
  • Most commonly damaged in the rear.

The most common reporting entities were:

  • Waymo LLC, owned by Google (62 crashes).
  • Transdev Alternative Services (34 crashes).
  • General Motors’ Cruise LLC (23 crashes).

Data in context

The NHTSA is the first to acknowledge shortcomings in the collected data. One flaw they group points out is that all of the crashes were categorized by what systems were equipped on the vehicle, not necessarily what systems were in use. That might cause crashes to be added to the wrong report.

Many variables can contribute to one manufacturer having more reported crashes than another. A company with a higher number of vehicles on the road, or vehicles that are on the road for many more miles, would likely report a larger number of crashes, though the incidence rate could be significantly smaller.

Manufacturers have significantly different capabilities for the detection of crashes, so an entity with more advanced telemetry systems can identify and report incidence more timely and accurately, increasing their reported crash number. Operating location can also play a significant role in the crash incidence.

Because of these issues, the standing general order is insufficient to determine whether these vehicles pose significant risks to the public and property, and the NHTSA advises against drawing conclusions just yet.

Ongoing NHTSA investigations

Last week, the NHTSA expanded an investigation started in August 2021 to “assess the performance of Tesla’s Autopilot system.” The primary investigation was prompted by incidents of Tesla vehicles crashing with first responders stopped on the road, often at the site of other accidents.

The investigation has now been upgraded to an in-depth engineering analysis. It is not limited to defects in the Tesla system but will pursue information to answer questions of whether these SAE Level 2 systems “exacerbate human factors or behavioral safety risks” due to a lack of driver engagement.

The results of this investigation will be applicable to a wide range of consumer automation controls. The government’s AV Test initiative provides information on autonomous vehicle testing, legislation and more to increase public awareness of the quickly expanding technology.