//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
On June 15, 2022, the National Highway Traffic Safety Administration (NHTSA) released a lot of collision data for L2 Advanced Driver Assistance Systems (ADAS) vehicles and Automated Driving System (ADS) vehicles, which provides L2, L3, and L4 data. A previous column looked at the ADS data.
This column looks at the L2 ADAS information from the NHTSA and will summarize the NHTSA ADAS information from the documents listed below. I also include some quarterly Tesla collision data that is available online.
The statistics cover July 2021 through May 15, 2022. The NHTSA plans to release similar data monthly. You can download several datasets directly from the NHTSA:
NHTSA L2 ADAS Data
The NHTSA released three data sets. The ADS, or L3 and L4, data includes a summary of information in a PDF file and detailed incident information in an Excel file. There are 130 incidents from July 2021 to May 2022. ADAS or L2 information is the second data set and is the focus of this column.
The third or “Other” data set has many incidents, but few have detailed information. It was also covered in the ADS–centric column.
The NHTSA provided a nine–page summary of the L2 ADAS data with several useful figures that are included in this column. The data covers 392 collisions from July 2021 to May 15, 2022.
There is also an Excel file with detailed information on all L2 ADAS collisions. The Excel file has 765 entries including 373 duplicates that covers the 392 crashes. The duplicate entries are due to multiple fillings per incident because of NHTAS rules. Most fillings are filled within one day and/or 10 days, plus updates when new information are available.
The spreadsheet entry has detailed collision information with up to 122 columns of data. The Data Element Definition file explains what each column means (linked above). Useful information before the crash, during the crash, and after the crash are requested, including vehicle movements, weather and lighting conditions, road conditions, location coordinates, collision severity, vehicle damage areas, addresses, and more. Personal and related information as well as company proprietary are redacted.
There are fields within the Data Element Definition file, however, that lists the ADAS information, but few entries have meaningful data. For instance, the ADAS version is one field, but the next field marks the ADAS version as confidential business information. Tesla and Ford redacted their ADAS version fields.
For example, the “Narrative” field may contain excellent information because it is the largest field at 4,000 characters and may contain a long description of a crash. Both Tesla and Ford, however, had this field redacted.
L2 ADAS crashes by reporting entities
The following figure summarizes the names of the vehicle companies reporting crashes. The data is from a figure on page five located in the NHTSA L2 ADAS summary report.
Tesla has the most L2 ADAS vehicles driving on U.S. roads and hence it is not a surprise that it has the most collision reports. My analysis of Tesla battery electric vehicle deliveries in the U.S. gives an estimate of 1.4 million L2 vehicles in use at the end of 2021 and is likely to top more than 2 million at the end of 2022. We do not know what portion of these vehicles are using the Tesla Autopilot and how often it is used.
Other data points are Tesla’s announcement of how many miles have been covered with Autopilot in use. In November 2016, Tesla said Autopilot covered 300 million miles, which increased to 1+ billion miles in November 2018. The last public data point was 3+ billion miles in February 2020.
If I scale this at a similar rate, current Tesla Autopilot cumulative miles would be around 6+ billion as the number of Tesla Autopilot vehicles have more than doubled since February 2020. These Autopilot numbers are on a worldwide basis, but the U.S. will account for well over 50%.
Additionally, the Tesla Autopilot can be used on highways and in city driving. Most other L2 ADAS systems can primarily be used on highways.
Honda has the second most collisions with 90, which is unexpected. The ADAS spreadsheet includes Honda and Acura models ranging from model year 2015 to 2022. It seems some of the collisions listed for Honda, however, are L1 ADAS vehicles.
L2 ADAS crashes by state
The NHTSA also provided statistics on which states had the most L2 ADAS collisions. The next figure is a copy of the NHTSA’s pictorial state summary. California is the overwhelming leader with 125 collisions, or 33% of total collisions. California had more collisions than Florida, Texas, New York, Pennsylvania, and Ohio combined.
Florida, Texas, and New York reported 34, 33, and 30 collisions, respectively. Pennsylvania and Ohio reported 13 and 12 collisions, respectively.
L2 ADAS crashes by injury severity
The next figure shows the severity of reported collisions. Unfortunately, 294 collisions (75%) of the total have no data available. In 46 cases there were no injuries, which is 47% of the crashes with severity response.
Minor injuries occurred in 19 collisions or 19% of severity response crashes. Moderate injuries contributed 22 collisions or 22%. Serious injuries occurred in five cases or 5%. There were six fatality collisions.
What did L2 ADAS vehicles collide with?
The next table shows NHTSA data on what the L2 ADAS vehicles collided with. The table has two data sets — total collisions from the NHTSA ADAS summary and Tesla collisions, which I extracted from the ADAS Excel file.
Total crashes with other vehicles reached 116 or nearly 30%. Tesla had 92 collisions with other vehicles, which is nearly 34%. Passenger car crashes reached 62 or 16% of the total collisions. Tesla has 51 collisions involved with passenger cars or nearly 19%. There were two crashes with First Responder Vehicles — both were by Tesla vehicles.
The number of crashes with fixed objects include poles/trees at 20 collisions and other fixed objects at 78 collision for a total of 98 collisions, which accounts for 25% of all collisions. Tesla’s equivalent data was 77 collisions or 28% of total crashes. These percentages seem on the high side. Does this mean that L2 vehicles have trouble recognizing fixed objects? And by extension, L3 and L4 may have similar issues?
The crashes with people and animals reached 14 or 3.6% of all collisions. Tesla had five crashes with people and animals (1.8%).
Tesla had a lower percentage of unknown crashes at 31% versus 37% for all companies. This is probably due to telematics systems, cameras, and rapid data extraction by Tesla.
L2 ADAS vehicle damage
The next table shows NHTSA data on L2 ADAS vehicle damage. The table has two data sets — total collisions from the NHTSA ADAS summary and Tesla collisions, which I extracted from the ADAS Excel file.
It is common that a vehicle will receive damage in multiple areas. Hence, the sums of the damage counts are higher than the total number of collisions. When I analyzed the Tesla data, I found 80 entries that did not have any damage information and should have been marked as unknown. This also changed the unknown count for all vehicles.
I have focused on the sum of the three front areas and three rear areas. I think the relative ratio between these two segments provides some information about the fault of the collision. The high crash percentage of rear area damage indicates low fault rate of the ADAS vehicle. The high collision percentage of front area damage indicates high fault of the ADAS vehicle.
The percentages of front and rear damages have been calculated with unknown crashes included and excluded. I think excluding unknown crashes is a better indicator of collision faults.
With unknowns included in the calculations, the damage of front areas account for 53.3% of all vehicles while damage for rear areas was 13.1%. The equivalent figures for Tesla were 26.2% for front areas and 21.7% for rear areas.
With unknowns excluded from the calculations, the damage of front areas added up to 70.6% of all vehicles while damage for rear areas was 17.4%. The corresponding data for Tesla was 44.8% for front areas and 37.1% for rear areas. This data can be found in the two last lines in the above table.
This data indicates that Tesla L2 ADAS vehicles might be performing better than other L2 ADAS vehicles in terms of who is at fault in the collisions. This is complicated, however, and there might be other factors at work.
There is a recent report that claims Tesla is not reporting all L2 ADAS crashes. This report says over 70% of all Tesla crashes involving L2 ADAS relate to Full Self–Driving software.
Tesla proprietary crash data
Tesla has released its own safety data on a quarterly basis since Q3 2018. The latest data is Q4 2021.
Tesla relied on the following methodology for its data collection:
- Tesla collects miles traveled by each vehicle with Autopilot active or in manual driving.
- Tesla also received crash information anytime a crash is reported from the fleet, which might include data about whether Autopilot was active at the time of impact.
- Tesla counts any crash in which Autopilot was deactivated within five seconds before impact and counts all crashes in which the incident alert indicated an airbag or other active restraint deployed. This correlates to nearly any crash at about 12 mph (20 kph) or above, depending on the crash forces generated.
- Over 35% of all Autopilot crashes occur when the Tesla vehicle is rear–ended by another vehicle.
Much more crash details are needed to compare the Tesla data with the NHTSA’s collision statistics. There is no data on how many crashes the Tesla fleet were involved in or how many miles were driven during each quarter.
From the online data, Tesla provide three measures of its crash and fire data:
- Average miles driven between crashes for all Tesla vehicles using manual ADAS.
- Average miles driven between crashes for all Tesla vehicles using Autopilot.
- Average miles driven between fires for all Tesla vehicles — yearly data only.
The next table is a summary of Tesla’s online data. There is no data for 2022 yet.
According to Tesla’s data, the average miles between crashes for Tesla Autopilot vehicles are much better than for Tesla vehicles with active safety or ADAS features. During 2021, Autopilot vehicles have 2x to 3x better crash data than ADAS vehicles. There is no additional data, however, to provide better perspectives on this trend.
For Q4 2021 crash comparisons, Tesla quotes NHTSA data that shows one automobile crash for every 484,000 miles driven. This data point is close to 2019 NHTSA statistics — total vehicles miles driven (3.262 billion) divided by total police reported crashes (6.76 million).
For fire comparisons, Tesla quotes data from the National Fire Protection Association and U.S. Department of Transportation that shows in the U.S. there is a vehicle fire for every 19 million miles traveled.
There is a lot of information from the NHTSA’s ADAS data, but there is a long way to go to get actionable statistics. We especially need vehicle miles travelled to get better analysis of both L2 ADASL2 and ADS improvement trends. Maybe the redacted data would add more useful information.
When monthly data becomes available, there will hopefully be less redacted data and better data in the 122 available fields of information. Less redacted data in the “Narrative” field and the ADS/ADAS version field would be especially useful.
Long term, we need NHTSA to gather three types of data as an aggregate for all auto OEMs and related service providers:
- Collisions by key categories: L2, L3, L4, and eventually L5. At least yearly and preferably quarterly.
- Miles driven by key category: L2, L3, L4, and eventually L5. At least yearly and preferably quarterly.
- Collision severity crash reports — including injuries and property damage–only crashes.
Aggregate data should be public data. Similar data for individual companies would be great but would likely meet strong resistance from most auto OEMs. Such data will allow us to track the improvements of ADAS and ADS technology to see what works best.