Road-safety research · Research Centre for Integrated Transport Innovation (rCITI), UNSW Sydney
Road-safety evidence that holds up when it is questioned
SafeFuture Lab builds statewide road-safety evidence from raw public data: traffic signals, traffic volumes, road geometry, speeds and crashes. You can trace every number we publish back to its source record and re-run it from the raw data, and the person who built the pipeline can defend it.
What the lab's work is
The lab works in six areas. Each one has a public description on our expertise pages, and there are case notes on the method, the numbers and the limits.
System integrity
Pipelines are gated and reproducible, limits are published, and estimates come with honest uncertainty. This sits under all the other work.
Statewide network evidence
Signal configurations, detector counts, directional traffic volumes, camera-based counts and speed-limit checks, built as one checked chain.
Spatial road risk
Intersection skew, vertical grade, curvature and network context, each joined to crash history.
Speed, evaluation and connected vehicles
Before/after evaluation, speed-zone studies and connected-vehicle methods, tested against instrumented research cars.
Crash data systems and analysis
Crash analysis and reporting systems, national fatality analysis, and trend explanations tested against gates set before the data arrive.
Verified AI and machine learning
AI that answers only what it can check against sources, and crash models that state their limits.
At a glance
These are aggregate figures from the lab's own pipelines, run on open data. Victoria was the first full build. The methods carry over to any state that publishes the same inputs.
Where government or industry has used it
We have not yet put up public evidence of a named government or industry use. We would rather tell you that than imply a use we cannot link to.
The Work relied on page lists each piece of work with a link to its public description. Under Evidence we are still gathering it lists the uses we are still documenting. Each will move up once there is a public record to link to.
How we work
- Source before action. Every input has a known, public source.
- Root cause before reply. When something goes wrong, we find out why before we answer.
- Every estimate ships with its uncertainty. Weak results are labelled by their interval, not hidden. See a worked example.