Driver-assist NHTSA crash data called limited, useful

WASHINGTON — Despite caveats in NHTSA's first-ever release of data on crashes linked to advanced driver-assistance systems and automated-driving systems, safety experts say the effort is still a useful step toward greater transparency.

The nation's top highway safety agency this week said it received reports of nearly 400 crashes involving driver-assist systems and 130 crashes involving fully automated vehicles since it began forcing automakers, suppliers and tech companies to disclose the data last June.

As automakers add more advanced driver-assistance features such as adaptive cruise control and lane-keeping assist to their vehicles, NHTSA said it wants to use the more timely data to better understand their role in crashes, whether they're caused by design flaws in the technology or misuse by drivers, and ensure their safety.

For AVs, which are not yet available to consumers but are being tested and deployed in a limited scale on public roads, the agency said it will evaluate how the vehicles interact with their surroundings and keep occupants safe.

"Safety needs transparency," said Phil Koopman, a Carnegie Mellon University professor who has been working in the AV safety field for more than two decades. "Without transparency, you don't get safety."

According to U.S. data, Tesla Inc. and American Honda Motor Co. — two automakers with driver-assist features currently under investigation by NHTSA — reported the most Level 2 advanced driver-assistance system crashes to the agency: 273 and 90, respectively. Subaru reported 10 crashes, followed by Ford Motor Co. with five. The remaining nine automakers reported four or fewer crashes.

Of the 392 total crashes involving advanced driver-assistance systems reported by the 12 automakers, six reports involved a death and five had serious injuries, NHTSA data showed. Injury severity in some cases may be unknown if it's not provided by a company's crash data source, the agency noted.

"I think what this data does, in a way, is it requires manufacturers who may have been hesitant to report these types of incidents, or who are not doing as great of a job reporting things to NHTSA in general … to get that data into the system," said Michael Brooks, acting executive director at the Center for Auto Safety. "It's an immediate update to the agency."

Still, NHTSA senior officials this week cautioned that the data is preliminary and lacks proper context such as the number of vehicles operating on U.S. roads with these technologies, total vehicle miles traveled during the reporting period and how the vehicles compare with those that don't have advanced driver-assistance or automated-driving systems.

"The data alone may raise more questions than they answer," NHTSA Administrator Steven Cliff warned reporters in a press call this week.

One crash can be counted multiple times and, in some cases, reporting companies may mistakenly classify the technology as an automated-driving system when it is actually driver-assist — and vice versa.

Companies also have varying capabilities related to crash data recording and a vehicle's ability to remotely transmit data to the manufacturer.

"I think it's fair to say that manufacturers who have better data recording and crash reporting systems built into their cars might be overrepresented in the NHTSA data," said Brooks. "It might look like Honda may be having a lot more problems than other manufacturers and that may be because they're simply reporting a higher percentage of the crashes."

Honda spokesman Chris Martin said about 6 million Acura and Honda vehicles in the U.S. are equipped with its AcuraWatch and Honda Sensing driver-assist systems and that "the population of vehicles that theoretically could be involved in a reportable event is much greater than the population of vehicles built by automakers with a less aggressive deployment strategy."

As automakers improve their crash-data recording capabilities across all vehicles, the data could become more closely aligned or show stronger trends, Brooks said. "But right now," he added, "it's probably well too early for that."

As is, the data cannot be used to compare system safety or crash rates among manufacturers, NHTSA said.

"There's no way you can tell if the number of crashes is a lot or how it compares to human drivers," said Todd Benoff, a partner in Alston & Bird's products liability practice group and a co-founder of the firm's connected and autonomous vehicles team. "If you just take the raw number of accidents, the only way you can get the analysis right is by blind luck."

NHTSA does not currently require companies subject to the order to report vehicle miles traveled or fleet size but could consider updating its crash-reporting requirements to include more contextual information.

In the meantime, the agency said it will continue to evaluate the crash reports and plans to release data updates monthly.

While the data is too limited to make "any sort of meaningful conclusions," Benoff said, there are still trends worth keeping an eye on.

Vehicles with advanced driver-assistance systems, for instance, "were most commonly damaged in the front, which means the human drivers are running into things," he noted.

Meanwhile, AVs "were most commonly damaged in the rear, which usually means that someone else did the crashing."

But, Benoff added, "This data was only ever intended to spot trends and possible issues to investigate."

NHTSA senior officials confirmed that the data had already been used to trigger new investigations and recalls, and inform ongoing safety probes.

Tesla's Autopilot driver-assist system is under escalated scrutiny after a series of crashes in the U.S. that resulted in more than a dozen injuries and one death. The agency this month upgraded its investigation of the system to an engineering analysis and later could seek a recall of an estimated 830,000 vehicles.

The crashes — now totaling 16 — involved Tesla vehicles with Autopilot engaged that were driven near first-responder scenes and subsequently struck one or more stopped emergency vehicles involved with those scenes.

NHTSA said it also closely reviewed 191 Tesla crashes involving a version of Autopilot, including the Full Self- Driving beta, during its preliminary evaluation. Some of those crashes were identified from NHTSA's crash reporting mandate.

"All the people tracking the race to autonomy aren't going to find this data as usual because that's not NHTSA's job," said Carnegie Mellon's Koopman. "NHTSA's job is to find defects and get them fixed, not be the scorekeeper."