Europe · Crash Reporting
BetaA deduplicated record of publicly reported crashes involving automated-driving systems across Europe. Unlike the US NHTSA SGO tracker, this beta is discovery-based: one real-world crash is counted once even when multiple media, operator or agency sources report it.
Coverage is not yet exhaustive. Records are added only when the event, automated-driving involvement, date/location and source trail can be verified well enough to avoid double counting.
Crash Details
Multiple-source matchTwo unmanned ParkShuttle vehicles collided head-on near a bridge after a communications problem and a supervisory release error allowed both vehicles into the same single-lane section. Connexxion's investigation said the obstacle-detection system reacted but the stopping distance was insufficient for two vehicles approaching one another. No passengers were aboard.
One row per real-world crash, not per article. Historical discovery searches cover European countries across the continent, including the United Kingdom, Norway and Switzerland. Media reports, operator statements, police/public-agency material, European project reports and academic evaluations describing the same event are merged into one canonical incident using compatible date/period, location, operator/manufacturer, vehicle, automation status and collision sequence. Exact dates and outcomes are left unknown when public sources do not establish them.
Beta limitation: This is a historical-search seed, not a guarantee of complete European coverage. Europe has no single public incident-level ADS crash feed comparable to the U.S. NHTSA SGO. Reporting regimes differ across European countries, and public availability varies substantially by country and language.
Historical search also finds reports that cannot yet be separated into individual incidents, or events that occurred while the vehicle was clearly in manual operation. They stay visible here so the research trail is not lost, but they are excluded from the incident count above.
Germany · By Dec 2021
Aggregate evidence only — individual crashes not yet separable
Operator reporting cited by Golem says the Monheim shuttle fleet had already been involved in multiple collisions, none attributed to the buses and none causing personal injury. The public article does not provide enough dates or event-level details to split these into defensible canonical records without further local-source research.
Austria · 2023-08-09
Excluded from canonical AV-crash count — manual operation
SURAAA's own statement confirms the shuttle damaged parked cars while being manually driven off the automated route during a routine parking maneuver. It explicitly attributes the event to human driving error and says no people were involved. Retained here for transparency, but not counted as an automated-driving crash.
United Kingdom · 2026-04-22
Not counted as a crash — manual mode; no collision reported
A Waymo test vehicle drove through a police cordon on Harlesden High Street. Waymo said the vehicle was in manual mode with a validation driver in full control. The driver was suspended pending investigation. No collision was reported, so this is retained as a reviewed safety-related incident rather than counted in the crash tracker.
The media tracker is only the first layer. These sources can improve completeness and provide official corroboration as access becomes available.
Manufacturers of covered fully automated vehicles must report safety-critical occurrences and periodic in-service monitoring information to type-approval/market-surveillance authorities and the European Commission.
UK pilot guidance requires operators to report safety-relevant occurrences, including collisions and traffic infractions. This creates a separate official reporting pathway for the UK after Brexit.
Examples include Sweden's STRADA, Norway's Statens vegvesen AV-testing permits and experience reports, Swiss FEDRO/cantonal records, and national type-approval authorities. These can corroborate or uncover crashes not visible in media searches.
Useful as a backstop for injury/fatal crashes, but standard public CARE fields do not directly identify ADS engagement, so matching requires external incident details.
Primary beta discovery layer. Multiple reports about one real-world event are merged into one canonical incident. Operator, police and public-agency statements are preferred when available.