🚦

Self-Driving Car Crash Tracker

3,056 crashes involving self-driving systems were reported to the National Highway Traffic Safety Administration between July 2021 and September 2026, by 35 companies. In 52% of them the self-driving vehicle was stopped or parked; 10% involved a reported injury. Read how to read these numbers before comparing companies.

Crashes reported: 3,056
Companies: 35
Vehicle stopped or parked: 52%
Injury reported: 10%; 298 crashes
Serious or fatal: 14; 3 fatal crashes
Data through: Sep 2026

Crashes reported per month

Self-driving crashes reported to NHTSA per month, Jul 2021–Sep 2026

Peak: March 2026, 145 crashes

View as a table

Month	Crashes
July 2021	12
August 2021	13
September 2021	11
October 2021	13
November 2021	14
December 2021	7
January 2022	11
February 2022	11
March 2022	16
April 2022	18
May 2022	15
June 2022	14
July 2022	15
August 2022	21
September 2022	11
October 2022	6
November 2022	9
December 2022	14
January 2023	8
February 2023	15
March 2023	22
April 2023	18
May 2023	22
June 2023	22
July 2023	29
August 2023	52
September 2023	34
October 2023	31
November 2023	11
December 2023	21
January 2024	23
February 2024	23
March 2024	28
April 2024	21
May 2024	40
June 2024	42
July 2024	40
August 2024	46
September 2024	58
October 2024	67
November 2024	70
December 2024	81
January 2025	78
February 2025	78
March 2025	77
April 2025	96
May 2025	110
June 2025	72
July 2025	66
August 2025	61
September 2025	67
October 2025	97
November 2025	112
December 2025	109
January 2026	118
February 2026	92
March 2026	145
April 2026	141
May 2026	118
June 2026	129
July 2026	120
August 2026	103
September 2026	12

Most of the climb follows the fleets: more robotaxis driving more miles in more cities means more crashes of every kind, including the ones where someone else hit a stopped car. The latest month or two fill in as late reports arrive.

By company

Company	Crashes	Injury reported	Serious or fatal	Stopped or parked	Reported
Waymo	2,382	213	11	58%	Jul 2021 – Sep 2026
Zoox	170	35	0	37%	Jul 2021 – Sep 2026
Cruise	156	27	3	31%	Jul 2021 – Nov 2024
Avride	84	3	0	21%	Oct 2023 – Aug 2026
May Mobility	48	4	0	33%	Jul 2021 – Aug 2026
Tesla	36	2	0	36%	Jul 2025 – Aug 2026
Motional	28	4	0	21%	May 2022 – Sep 2026
Aurora	24	1	0	0%	Mar 2023 – Aug 2026
Argo AI	20	3	0	60%	Jul 2021 – Oct 2022
Nuro	15	1	0	27%	Jan 2024 – Jul 2026
Beep	14	0	0	14%	Aug 2021 – Jan 2026
WeRide	13	1	0	54%	Jul 2021 – Mar 2026
Volkswagen ADMT	10	0	0	80%	Jan 2024 – Jun 2024
Kodiak	6	0	0	0%	Aug 2022 – Mar 2025
Pony.ai	6	0	0	33%	Oct 2021 – Sep 2023
Mercedes-Benz	5	2	0	20%	Mar 2022 – Aug 2023
MOIA	5	0	0	0%	Mar 2026 – Aug 2026
Torc Robotics	4	0	0	0%	Jul 2022 – Aug 2026
Baidu Apollo	3	0	0	0%	Jun 2023 – Oct 2023
Apple	3	0	0	33%	Sep 2021 – Sep 2023
Gatik	3	0	0	0%	Jun 2026 – Aug 2026
Colorado School of Mines	2	0	0	100%	Oct 2021
Ghost Autonomy	2	0	0	50%	Aug 2023 – Jan 2024
International Motors (Navistar)	2	0	0	0%	Nov 2021 – Aug 2026
NVIDIA	2	0	0	50%	Nov 2022 – Feb 2025
PlusAI	2	0	0	0%	May 2025 – Apr 2026
Stack AV	2	0	0	0%	Sep 2025 – Apr 2026
Waabi	2	1	0	0%	Aug 2026
AutoX Technologies	1	0	0	0%	Feb 2025
First Transit	1	0	0	0%	Aug 2022
NAVYA	1	0	0	0%	Sep 2021
Ohmio	1	1	0	0%	Aug 2025
Robotic Research	1	0	0	0%	Jan 2023
TuSimple	1	0	0	0%	Apr 2022
VisLab (Ambarella)	1	0	0	0%	Dec 2023

Sorted by crashes, which mostly sorts by miles driven: the companies at the top run the largest driverless fleets. Stopped or parked is the share of each company's crashes in which its vehicle was not moving.

Most serious injury reported

Fatal0.1% · 3 crashes
Serious injury0.4% · 11 crashes
Moderate injury1.2% · 36 crashes
Minor injury8.1% · 248 crashes
No injuries reported88.5% · 2,706 crashes
Unknown1.7% · 52 crashes

The most serious injury any report of the crash alleges, to anyone involved: people in the self-driving vehicle, in other vehicles, on bikes or on foot.

What the self-driving vehicle collided with

Another vehicle83.3% · 2,546 crashes
Other8.5% · 261 crashes
Fixed object4.5% · 138 crashes
Cyclist1.7% · 52 crashes
Animal0.9% · 29 crashes
Scooter, skateboard or other road user0.7% · 20 crashes
Pedestrian0.3% · 10 crashes

Who could take the wheel

No one on board to drive72.3% · 2,209 crashes
Safety driver on board23.1% · 706 crashes
Remote operator only4.2% · 129 crashes
Unknown0.3% · 9 crashes
Private owner driving0.1% · 3 crashes

No one on board to drive means a fully driverless ride; a safety driver means a test or early service with an employee ready to take over.

Where they happened

San Francisco, CA 1,128 crashes
Los Angeles, CA 390 crashes
Austin, TX 287 crashes
Phoenix, AZ 271 crashes
Tempe, AZ 126 crashes
Atlanta, GA 88 crashes
Dallas, TX 75 crashes
Las Vegas, NV 75 crashes
Scottsdale, AZ 67 crashes
Miami, FL 40 crashes
Houston, TX 35 crashes
Santa Monica, CA 32 crashes

The cities at the top are where the largest robotaxi services run, which is where the most driverless miles are driven.

How to read these numbers

A count is not a rate. A company that drives more miles reports more crashes. NHTSA does not publish the miles behind these reports, so nothing here is divided by distance, and a longer bar does not mean a less safe car.
Being in a crash is not causing it. The reports do not assign fault, and many of these crashes happened while the self-driving vehicle was stopped or parked, often hit from behind.
Companies file these reports themselves. NHTSA publishes them as filed: they may be incomplete or unverified, injuries are as alleged, and some details are withheld as confidential business information.
One crash, one row. A crash can get several reports, from updates or from more than one company (an operator and its fleet partner); we count it once, using NHTSA's own incident ID.
The rules changed in June 2025. NHTSA amended what must be reported and when, so counts before and after that month are not directly comparable.
Driver assistance is not here. Crashes with systems like Tesla Autopilot or GM Super Cruise in customers' cars are a separate NHTSA dataset with different rules. 15 reports filed in the self-driving file that name a driver-assistance system are left out.