//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
On June 15, 2022, the Nationwide Freeway Visitors Security Administration (NHTSA) launched quite a lot of collision information for L2 Superior Driver Help Programs (ADAS) automobiles and Automated Driving System (ADS) automobiles, which offers L2, L3, and L4 information. A earlier column regarded on the ADS information.
This column seems to be on the L2 ADAS info from the NHTSA and can summarize the NHTSA ADAS info from the paperwork listed under. I additionally embrace some quarterly Tesla collision information that’s accessible on-line.
The statistics cowl July 2021 by means of Could 15, 2022. The NHTSA plans to launch comparable information month-to-month. You may obtain a number of datasets instantly from the NHTSA:
NHTSA L2 ADAS Information
The NHTSA launched three information units. The ADS, or L3 and L4, information features a abstract of knowledge in a PDF file and detailed incident info in an Excel file. There are 130 incidents from July 2021 to Could 2022. ADAS or L2 info is the second information set and is the main target of this column.
The third or “Different” information set has many incidents, however few have detailed info. It was additionally coated within the ADS–centric column.
The NHTSA supplied a 9–web page abstract of the L2 ADAS information with a number of helpful figures which might be included on this column. The information covers 392 collisions from July 2021 to Could 15, 2022.
There’s additionally an Excel file with detailed info on all L2 ADAS collisions. The Excel file has 765 entries together with 373 duplicates that covers the 392 crashes. The duplicate entries are attributable to a number of fillings per incident due to NHTAS guidelines. Most fillings are crammed inside at some point and/or 10 days, plus updates when new info can be found.
The spreadsheet entry has detailed collision info with as much as 122 columns of knowledge. The Information Aspect Definition file explains what every column means (linked above). Helpful info earlier than the crash, throughout the crash, and after the crash are requested, together with car actions, climate and lighting circumstances, highway circumstances, location coordinates, collision severity, car harm areas, addresses, and extra. Private and associated info in addition to firm proprietary are redacted.
There are fields throughout the Information Aspect Definition file, nonetheless, that lists the ADAS info, however few entries have significant information. As an illustration, the ADAS model is one area, however the subsequent area marks the ADAS model as confidential enterprise info. Tesla and Ford redacted their ADAS model fields.
For instance, the “Narrative” area could comprise glorious info as a result of it’s the largest area at 4,000 characters and will comprise a protracted description of a crash. Each Tesla and Ford, nonetheless, had this area redacted.
L2 ADAS crashes by reporting entities
The next determine summarizes the names of the car firms reporting crashes. The information is from a determine on web page 5 situated within the NHTSA L2 ADAS abstract report.
Tesla has essentially the most L2 ADAS automobiles driving on U.S. roads and therefore it’s not a shock that it has essentially the most collision experiences. My evaluation of Tesla battery electrical car deliveries within the U.S. provides an estimate of 1.4 million L2 automobiles in use on the finish of 2021 and is more likely to prime greater than 2 million on the finish of 2022. We have no idea what portion of those automobiles are utilizing the Tesla Autopilot and the way typically it’s used.
Different information factors are Tesla’s announcement of what number of miles have been coated with Autopilot in use. In November 2016, Tesla stated Autopilot coated 300 million miles, which elevated to 1+ billion miles in November 2018. The final public information level was 3+ billion miles in February 2020.
If I scale this at the same price, present Tesla Autopilot cumulative miles could be round 6+ billion because the variety of Tesla Autopilot automobiles have greater than doubled since February 2020. These Autopilot numbers are on a worldwide foundation, however the U.S. will account for properly over 50%.
Moreover, the Tesla Autopilot can be utilized on highways and in metropolis driving. Most different L2 ADAS methods can primarily be used on highways.
Honda has the second most collisions with 90, which is surprising. The ADAS spreadsheet contains Honda and Acura fashions starting from mannequin yr 2015 to 2022. It appears a few of the collisions listed for Honda, nonetheless, are L1 ADAS automobiles.
L2 ADAS crashes by state
The NHTSA additionally supplied statistics on which states had essentially the most L2 ADAS collisions. The following determine is a duplicate of the NHTSA’s pictorial state abstract. California is the overwhelming chief with 125 collisions, or 33% of complete collisions. California had extra collisions than Florida, Texas, New York, Pennsylvania, and Ohio mixed.
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 harm severity
The following determine exhibits the severity of reported collisions. Sadly, 294 collisions (75%) of the entire don’t have any information accessible. In 46 circumstances there have been no accidents, which is 47% of the crashes with severity response.
Minor accidents occurred in 19 collisions or 19% of severity response crashes. Reasonable accidents contributed 22 collisions or 22%. Severe accidents occurred in 5 circumstances or 5%. There have been six fatality collisions.
What did L2 ADAS automobiles collide with?
The following desk exhibits NHTSA information on what the L2 ADAS automobiles collided with. The desk has two information units — complete collisions from the NHTSA ADAS abstract and Tesla collisions, which I extracted from the ADAS Excel file.
Complete crashes with different automobiles reached 116 or almost 30%. Tesla had 92 collisions with different automobiles, which is almost 34%. Passenger automotive crashes reached 62 or 16% of the entire collisions. Tesla has 51 collisions concerned with passenger vehicles or almost 19%. There have been two crashes with First Responder Autos — each had been by Tesla automobiles.
The variety of crashes with fastened objects embrace poles/bushes at 20 collisions and different fastened objects at 78 collision for a complete of 98 collisions, which accounts for 25% of all collisions. Tesla’s equal information was 77 collisions or 28% of complete crashes. These percentages appear on the excessive facet. Does this imply that L2 automobiles have hassle recognizing fastened objects? And by extension, L3 and L4 could have comparable points?
The crashes with folks and animals reached 14 or 3.6% of all collisions. Tesla had 5 crashes with folks and animals (1.8%).
Tesla had a decrease proportion of unknown crashes at 31% versus 37% for all firms. That is most likely attributable to telematics methods, cameras, and fast information extraction by Tesla.
L2 ADAS car harm
The following desk exhibits NHTSA information on L2 ADAS car harm. The desk has two information units — complete collisions from the NHTSA ADAS abstract and Tesla collisions, which I extracted from the ADAS Excel file.
It’s common {that a} car will obtain harm in a number of areas. Therefore, the sums of the harm counts are greater than the entire variety of collisions. After I analyzed the Tesla information, I discovered 80 entries that didn’t have any harm info and may have been marked as unknown. This additionally modified the unknown depend for all automobiles.
I’ve targeted on the sum of the three entrance areas and three rear areas. I believe the relative ratio between these two segments offers some details about the fault of the collision. The excessive crash proportion of rear space harm signifies low fault price of the ADAS car. The excessive collision proportion of entrance space harm signifies excessive fault of the ADAS car.
The odds of entrance and rear damages have been calculated with unknown crashes included and excluded. I believe excluding unknown crashes is a greater indicator of collision faults.
With unknowns included within the calculations, the harm of entrance areas account for 53.3% of all automobiles whereas harm for rear areas was 13.1%. The equal figures for Tesla had been 26.2% for entrance areas and 21.7% for rear areas.
With unknowns excluded from the calculations, the harm of entrance areas added as much as 70.6% of all automobiles whereas harm for rear areas was 17.4%. The corresponding information for Tesla was 44.8% for entrance areas and 37.1% for rear areas. This information might be discovered within the two final strains within the above desk.
This information signifies that Tesla L2 ADAS automobiles is perhaps performing higher than different L2 ADAS automobiles by way of who’s at fault within the collisions. That is sophisticated, nonetheless, and there is perhaps different elements at work.
There’s a latest report that claims Tesla isn’t reporting all L2 ADAS crashes. This report says over 70% of all Tesla crashes involving L2 ADAS relate to Full Self–Driving software program.
Tesla proprietary crash information
Tesla has launched its personal security information on a quarterly foundation since Q3 2018. The most recent information is This fall 2021.
Tesla relied on the next methodology for its information assortment:
- Tesla collects miles traveled by every car with Autopilot lively or in guide driving.
- Tesla additionally acquired crash info anytime a crash is reported from the fleet, which could embrace information about whether or not Autopilot was lively on the time of impression.
- Tesla counts any crash by which Autopilot was deactivated inside 5 seconds earlier than impression and counts all crashes by which the incident alert indicated an airbag or different lively restraint deployed. This correlates to almost any crash at about 12 mph (20 kph) or above, relying on the crash forces generated.
- Over 35% of all Autopilot crashes happen when the Tesla car is rear–ended by one other car.
Far more crash particulars are wanted to match the Tesla information with the NHTSA’s collision statistics. There is no such thing as a information on what number of crashes the Tesla fleet had been concerned in or what number of miles had been pushed throughout every quarter.
From the web information, Tesla present three measures of its crash and fireplace information:
- Common miles pushed between crashes for all Tesla automobiles utilizing guide ADAS.
- Common miles pushed between crashes for all Tesla automobiles utilizing Autopilot.
- Common miles pushed between fires for all Tesla automobiles — yearly information solely.
The following desk is a abstract of Tesla’s on-line information. There is no such thing as a information for 2022 but.
In response to Tesla’s information, the typical miles between crashes for Tesla Autopilot automobiles are significantly better than for Tesla automobiles with lively security or ADAS options. Throughout 2021, Autopilot automobiles have 2x to 3x higher crash information than ADAS automobiles. There is no such thing as a extra information, nonetheless, to supply higher views on this development.
For This fall 2021 crash comparisons, Tesla quotes NHTSA information that exhibits one car crash for each 484,000 miles pushed. This information level is near 2019 NHTSA statistics — complete automobiles miles pushed (3.262 billion) divided by complete police reported crashes (6.76 million).
For fireplace comparisons, Tesla quotes information from the Nationwide Hearth Safety Affiliation and U.S. Division of Transportation that exhibits within the U.S. there’s a car fireplace for each 19 million miles traveled.
Abstract
There’s quite a lot of info from the NHTSA’s ADAS information, however there’s a lengthy technique to go to get actionable statistics. We particularly want car miles travelled to get higher evaluation of each L2 ADASL2 and ADS enchancment tendencies. Possibly the redacted information would add extra helpful info.
When month-to-month information turns into accessible, there’ll hopefully be much less redacted information and higher information within the 122 accessible fields of knowledge. Much less redacted information within the “Narrative” area and the ADS/ADAS model area could be particularly helpful.
Long run, we’d like NHTSA to collect three varieties of information as an mixture for all auto OEMs and associated service suppliers:
- Collisions by key classes: L2, L3, L4, and finally L5. At the least yearly and ideally quarterly.
- Miles pushed by key class: L2, L3, L4, and finally L5. At the least yearly and ideally quarterly.
- Collision severity crash experiences — together with accidents and property harm–solely crashes.
Mixture information ought to be public information. Comparable information for particular person firms could be nice however would probably meet robust resistance from most auto OEMs. Such information will enable us to trace the enhancements of ADAS and ADS know-how to see what works greatest.
