About

White Box Autonomy runs the Global AV Event Observatory: a structured, openly sourced catalogue of events involving automated-driving systems.

What we record

Each entry is an event — a record that something happened involving a putative automated vehicle. An event links to one or more sources — the original posts, reporting or documentation — which are its evidence, alongside a short description written by White Box Autonomy. We record both failures and the successful handling of difficult situations, so the catalogue is not limited to things that went wrong.

How events enter the Observatory

Curated events come from public evidence. Some are found through automated discovery: two monitored sources on X — the curator’s bookmarks and White Box Autonomy’s own posts — are checked once a day, and AI helps screen what they contain and decide which items deserve a curator’s attention. AI does not establish event facts, write event descriptions or publish anything. A curator watches the evidence, checks for duplicates, writes the event’s description and decides whether to publish it. The evidence is always the original sources, linked from every event.

Source-derived events are created differently: through a validated, deterministic transformation of an identified structured source dataset (currently NHTSA’s Standing General Order incident reports), without individual curator review or AI involvement. Both pathways are explained on the methodology page.

Principles

• AI discovers and structures information; AI itself is never evidence.
• Every machine-discovered event has traceable source provenance.
• Missing facts are left unknown — we do not infer them.
• The involvement of an automated vehicle is not the same as its being the cause.
• Verification status is always explicit.
• AI-generated and curator-written text remain distinguishable.
• Publication is a curator’s decision. AI never publishes an event.

Current status

The Observatory currently offers this set of information pages, a browsable and filterable list of events, an interactive map for exploring their rough spatial distribution, individual event pages and a growing multi-source evidence base. Curated events are added through automated discovery from the two monitored sources described above, followed by curator review and publication; source-derived records are added from regulatory data. Broader searching is not yet active, and candidates are reviewed by a curator over time. Everything else in the long-term design — scenario extraction, benchmarking, contributor accounts, semantic search — comes later.

The methods behind discovery, provenance, and verification status are described on the methodology page.

Who runs this

White Box Autonomy is developed and maintained by the M2CAT Lab at The University of Queensland, with support from the Australian Research Council Future Fellowship (FT250100337) and The University of Queensland.

The views expressed on this website are those of the authors and do not necessarily reflect the views of The University of Queensland, the Australian Research Council, or the Australian Government.

Contact and corrections

To report an error, request a correction, or raise a concern about a record or its source material, contact whiteboxautonomy@uq.edu.au.