Public safety record
When a robot goes wrong,
someone files a record.
Robots are on public roads, in operating theatres, on factory floors and in people's homes. When one goes wrong, a record is usually created somewhere — by a workplace inspector, a road-safety regulator, a medical-device registry, or a recall notice. Those records sit in different countries, in different formats, behind different search forms.
Robolist pulls them from each regulator, normalises them into one shape, and keeps every record linked to the filing it came from. This is a standing collection, not a one-off report — it grows each time we sweep another registry, and the count above moves with it.
2,806
records, each linked to its source
14 registries · 1987–2026
last added 2026-09-11
95% of these records come from two American registries — FDA MAUDE and NHTSA. That shape is a fact about which governments publish, not about where robots fail.
Three rules for reading this page
No league table
Count and rate rank these operators in opposite orders.
Reporting volume tracks how much a company operates and how diligently it files. A ranking built on it would reward staying quiet.
A report is not a verdict
A car rear-ended at a red light files the same report as the car that hit it.
The FDA states that an adverse-event report does not establish the device caused the outcome. Only recalls and completed workplace investigations attribute a cause — those are marked.
Deaths are never one number
Five registries record a death. They are counting different things.
Each figure below is shown against the registry it came from, and this page never adds them together.
Where the records come from
Eight public registries in four jurisdictions. Each keeps its own definition of what is worth recording, which is why the counts below are not comparable with one another.
- FDA adverse event reportsFDA1,375317 diedReports involving surgical robots. A report is an allegation, not a finding.2002–2026
- NHTSA Standing General OrderNHTSA1,3031 diedCrash reports that driverless-vehicle operators are legally required to file.2025–2026
- OSHA accident investigationsOSHA6031 diedUS workplace accidents where an investigator found a robot in the chain of events.1987–2024
- Press and public records242 diedIndividually checked reports from news and regulator announcements.2008–2026
- Ca recallsHealth Canada12Public records collected under this source.2019–2026
- EU Safety Gate alertsEU Safety Gate11European rapid alerts on dangerous non-food consumer products.2010–2025
- Japan MHLW workplace recordsMHLW52 diedJapanese occupational accident records naming a robot.
- CPSC recallsCPSC4US consumer product recalls on robotic products.2022–2026
- Uk opssOPSS3Public records collected under this source.2022–2025
- Au acccACCC Product Safety3Public records collected under this source.2023–2026
- Cn samrdprcSAMR Defective Product Recall Technical Centre2Public records collected under this source.2025–2026
- Cn samrSAMR (State Administration for Market Regulation)2Public records collected under this source.2021–2024
- Tw molTaiwan MOL Occupational Safety and Health Administration11 diedPublic records collected under this source.2023–2023
- Jp niteNITE1Public records collected under this source.2021–2021
They do not count the same kind of event
This is what “not comparable” means in practice. An OSHA file is opened only when someone is killed or seriously hurt. A NHTSA crash report is filed for a dented bumper with nobody in the car. Setting those two counts side by side measures paperwork, not harm.
- FDA MAUDE23% involve injury or death1,375
317 death reported · 1,058 not graded by the source
- NHTSA10% involve injury or death1,303
1 death reported · 8 serious injury · 118 minor injury · 1,103 property damage only · 51 no injury reported · 22 not graded by the source
- OSHA100% involve injury or death60
31 death reported · 29 serious injury
- Press records21% involve injury or death24
2 death reported · 1 serious injury · 2 minor injury · 5 property damage only · 10 no injury reported · 4 not graded by the source
- Ca recalls8% involve injury or death12
- EU Safety Gate0% involve injury or death11
- Japan MHLW100% involve injury or death5
2 death reported · 3 serious injury
- CPSC0% involve injury or death4
- Au accc0% involve injury or death3
- Uk opss0% involve injury or death3
- Cn samr0% involve injury or death2
- Cn samrdprc0% involve injury or death2
- Jp nite0% involve injury or death1
- Korea KATS0% involve injury or death1
- Tw mol100% involve injury or death1
1 death reported
Death reportedSerious injuryMinor injuryProperty damage onlyNot graded by the source
They do not cover the same years
Each registry spans a different period, and some of them do not overlap at all — so a year-on-year total across all of them would measure when we swept each source, not when anything happened.
- OSHA1987–2024
- FDA MAUDE2002–2026
- Press records2008–2026
- EU Safety Gate2010–2025
- Ca recalls2019–2026
- Cn samr2021–2024
- Jp nite2021–2021
- CPSC2022–2026
- Uk opss2022–2025
- Au accc2023–2026
- Tw mol2023–2023
- NHTSA2025–2026
- Cn samrdprc2025–2026
No dates recorded for Japan MHLW.
How serious were they
This is the breakdown a headline count hides. Most filed records describe damage to a vehicle and nobody hurt.
- Death reported
five registries · never summed
317 FDA MAUDE31 OSHA2 Press records2 Japan MHLW1 NHTSA1 Tw mol - Serious injury41
- Minor injury121
- Recall26
- Property damage only1,112
- No injury reported62
- Not graded by the source1,090
Why there is no league table
A crash count on its own is unreadable without knowing how far the fleet drove. Ranking the same operators by count and then by rate produces opposite orders — this is that comparison, drawn.
6.0×lower reporting rate for the operator with the most reports
Waymo has filed 1,117 crash reports — more than every other driverless operator combined — against 220.6M published driverless miles (as of 2026-03-31), which is 5.06 per million. Zoox has filed 61 against 2.0M miles — 30.5 per million. Counting incidents ranks these two in the opposite order to measuring them.
Why some operators are absent
A rate needs a denominator big enough to mean something, so we only divide when an operator has published at least 1M miles. Aurora Innovation, Gatik are below it — one has filed a single report against 10,000 published miles, which arithmetic turns into “100 per million”, a figure built from one event that would be irresponsible to print.
Operators that publish no mileage
No published mileage found for Avride, Tesla, Nuro, Inc., WeRide, so no rate is shown for them. That is a statement about disclosure, not about safety. These are cumulative totals that are not perfectly time-aligned, and a filed report says a collision happened, not who caused it.
Every exposure figure we could verify
Very few operators publish how much they run. Each figure below is shown with the sentence it came from, so you can check it against the filing yourself.
- Waymo14 citiesas of 2026-09-01
“marking 14 cities where we're providing riders fully autonomous trips”
Source ↗ - WeRide3,400 vehiclesas of 2026-07-31
“our global L4 fleet comprised approximately 3,400 vehicles, including more than 1,800 robotaxis”
Source ↗ - Intuitive Surgical, Inc.11,710 systemsas of 2026-06-30
“The Company grew its da Vinci surgical system installed base to 11,710 systems as of June 30, 2026”
Source ↗ - Aurora Innovation440,000 milesas of 2026-06-30
“the Aurora Driver completed nearly 440,000 driverless miles since launch through the end of June”
Source ↗ - Serve Robotics Inc.1,800,000 deliveriesas of 2026-05-07
“1.8M Sidewalk + Hospital deliveries To date”
Source ↗ - Starship Technologies3,000 robotsas of 2026-04-28
“more than 3,000 autonomous robots”
Source ↗ - Starship Technologies10,000,000 deliveriesas of 2026-04-28
“Starship Technologies Passes 10 Million Deliveries”
Source ↗ - Waymo220,600,000 milesas of 2026-03-31
“Through March 2026, Waymo has driven 220.6M rider-only miles without a human driver”
Source ↗ - May Mobility1,100,000 milesas of 2026-03-24
“Since 2017, May Mobility has successfully given more than half a million autonomous rides and driven more than 1.1 million miles across 24 locations in the U.S. and Japan.”
Source ↗ - May Mobility24 locationsas of 2026-03-24
“across 24 locations in the U.S. and Japan”
Source ↗ - Zoox2,000,000 milesas of 2026-03-24
“In less than a year since our first launch in Las Vegas, we have driven nearly two million autonomous miles in our robotaxi and carried over 350,000 riders.”
Source ↗ - Gatik10,000 milesas of 2026-01-27
“It has completed over 10,000 driverless miles on public roads.”
Source ↗ - Serve Robotics Inc.2,000 robotsas of 2025-12-31
“2,000 Serve robots on sidewalks”
Source ↗
Workplace losses, by industry
OSHA records the employer, not the robot's manufacturer — the make is almost never named. So these are attributed only to an industry, and they never appear on any company's page. For anyone pricing risk in a plant, the industry is the useful part anyway.
- Manufacturing37 · 17 fatal
- Consumer / household25
- Not coded in the source23 · 15 fatal
- Medical9
- Accommodation & food2
- Professional services1 · 1 fatal
- Utilities1 · 1 fatal
- Retail1
- Transport & warehousing1
fatalother investigations
Individual records (30 of 2,806)
Deaths, serious injuries and recalls — the records that are meaningful one at a time. Text is excerpted verbatim from the registry and never rewritten; every card links to the original so you can read it and disagree with us.
- Death reported2026-06-30FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2026-06-22FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Serious injury2026-06NHTSA
On June [XXX], 2026 at 2:53 AM MT a Waymo Autonomous Vehicle ("Waymo AV") operating in Scottsdale, California was in a collision involving an SUV on [XXX] near [XXX].…
cause not establishedRead the record ↗ - Serious injury2026-05NHTSA
On May [XXX], 2026 at 3:59 PM MT a Waymo Autonomous Vehicle ("Waymo AV") operating in Phoenix, Arizona was in a collision involving a pickup truck on [XXX] at [XXX].…
cause not establishedRead the record ↗ - Serious injury2026-05NHTSA
On May [XXX], 2026 at 6:44 AM PT a Waymo Autonomous Vehicle ("Waymo AV") operating in Los Angeles, California was in a collision involving a pickup truck on [XXX] near {XXX}.…
cause not establishedRead the record ↗ - Recall2026-04-09CPSC
Wybotics Recalls Robotic Pool Vacuums Due to Burn and Fire Hazards — Hazard: The lithium-ion battery in the recalled pool vacuums can overheat, posing burn and fire hazards to consumers.…
cause attributed by sourceRead the record ↗ - Serious injury2026-04NHTSA
On April [XXX], 2026 at 9:24 PM PT a Waymo Autonomous Vehicle ("Waymo AV") operating in San Francisco, California was in a collision involving an SUV on [XXX] near [XXX].…
cause not establishedRead the record ↗ - Death reported2026-03-23FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2026-02-20Press records
A judge has denied Tesla's request to overturn a $243 million jury verdict that held the automaker partially responsible for a fatal crash involving its Autopilot driver assistance system.
cause not establishedRead the record ↗ - Serious injury2026-02NHTSA
On February [XXX], 2026 at 11:06 PM PT a Waymo Autonomous Vehicle ("Waymo AV") operating in San Francisco, California was in a collision involving a van on [XXX] near [XXX].…
cause not establishedRead the record ↗ - Death reported2026-01-29FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2026-01-26FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2025-11-03FDA MAUDE
Monarch Platform — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome. Reported narrative: IT WAS REPORTED THAT DURING A ROBOTICALLY ASSISTED BRONCHOSCOPY PROCEDURE, AS THE PHYSICIAN WAS TAKING…
cause not establishedRead the record ↗ - Serious injury2025-11NHTSA
On November [XXX], 2025 at 12:02 AM MT a Waymo Autonomous Vehicle ("Waymo AV") operating in Scottsdale, Arizona was in a collision involving a passenger of the Waymo AV on [XXX] near [XXX]. The Waymo AV was travelling northbound in the left lane of [XXX].…
cause not establishedRead the record ↗ - Death reported2025-10-15FDA MAUDE
ExcelsiusGPS — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome. Reported narrative: IT WAS REPORTED THAT AT THE HOSPITAL THEIR ELDERLY PATIENT WITH A THORACIC FRACTURE.…
cause not establishedRead the record ↗ - Serious injury2025-10NHTSA
On October [XXX], 2025 at 9:55 AM EDT a Waymo Autonomous Vehicle ("Waymo AV") operating in Atlanta, Georgia was in a collision involving a passenger car on [XXX] at [XXX].…
cause not establishedRead the record ↗ - Death reported2025-09NHTSA
On September [XXX], 2025 at 1:21 AM PT a Waymo Autonomous Vehicle ("Waymo AV") operating in Tempe, Arizona was in a collision involving a motorcycle on [XXX] at [XXX].…
cause not establishedRead the record ↗ - Death reported2025-08-01Press records
Last August, a jury awarded a $243 million verdict against Tesla over its culpability in a 2019 fatal crash in Florida that killed Naibel Benavides and critically injured Dillon Angulo.…
cause not establishedRead the record ↗ - Serious injury2025-07NHTSA
On July [XXX], 2025 at 1:57 PM PT a Waymo Autonomous Vehicle ("Waymo AV") operating in Santa Clara, California was in a collision involving an SUV on [XXX] near [XXX].…
cause not establishedRead the record ↗ - Death reported2025-04-22FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Recall2025-03-20CPSC
Aiper Seagull Pro Cordless Robotic Pool Vacuum Cleaners Recalled Due to Burn and Fire Hazards; Distributed by Shenzhen Aiper Intelligent Co.…
cause attributed by sourceRead the record ↗ - Death reported2025-03-06FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2025-03-01FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2025-02-03FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2025-01-13FDA MAUDE
Monarch Platform — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome. Reported narrative: IT WAS REPORTED THAT THE PATIENT CODED DURING PROCEDURE, THE PHYSICIAN HAD CONDUCTED 4-BIOPSIES AND ON THE 5TH…
cause not establishedRead the record ↗ - Death reported2025-01-10FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome. Reported narrative: DRS. (B)(6), PLEASE NOTE, MS.…
cause not establishedRead the record ↗ - Death reported2024-12-02OSHA
At 10:35 a.m. on December 2, 2024, Employee #1 was performing maintenance on a sensor inside a robotic enclosure. As he entered the area, the robot was stopped by entrance sensors.…
cause attributed by sourceRead the record ↗ - Death reported2024-10-07FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2024-06-27FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗ - Death reported2024-04-18FDA MAUDE
da Vinci Surgical System — Death report filed with FDA (MAUDE). A MAUDE report does not establish that the device caused the outcome.…
cause not establishedRead the record ↗
Where the robots are, and where the record is
The most useful thing this dataset can tell someone pricing risk is mostly about absence. These are the countries with the most robot makers on Robolist, against the number of incident records the registries in that jurisdiction actually produce.
- United States482 makers2,742 records4 registries
- China374 makersno public registry
- Japan78 makers5 records1 registry
- Germany73 makersno public registry
- South Korea51 makersno public registry
- France42 makersno public registry
- Canada41 makersno public registry
- Switzerland38 makersno public registry
- United Kingdom36 makersno public registry
- India31 makersno public registry
- Italy26 makersno public registry
- Taiwan22 makersno public registry
10 of these 12 countries have no public robot-incident registry we have been able to read — including China, Germany, South Korea. Their absence from every chart on this page is a fact about disclosure, not about safety. If you are pricing risk in one of them, this record cannot help you, and no other public record we know of can either.
What this is missing
Almost every record here is American, because the United States publishes this kind of data and most countries do not. Workplace robot accidents outside the US, Chinese and Gulf robotaxi incidents, and European industrial cases are largely absent — not because they did not happen, but because we have not yet found them in a public registry. Read an absence on this page as a gap in the public record, never as evidence that nothing occurred.
If we have a record wrong
These records are collected from public registries and reproduced as they were filed; we do not independently investigate them and we make no finding about who was at fault. Every entry links to the regulator that published it, so the original always outranks our copy. If a record is misattributed or out of date, the correction form on the relevant company's page reaches us directly, and how we handle corrections is set out in our transparency policy.