Observed driving envelope: one driver's band on one UK street, and a driving model's paths through it
One driver, twelve passes, one street: the band our driver kept to; too few passes to calibrate.
Loading the scored paths.
Measured: rig passes, LiDAR ground. Model: bench outputs, 2 Oct 2026. Street: open data.
- Wrong turn side (Highway Code 179 to 183): 3 of 19 (5.5 to 37.6 %) samples with route text, 4 of 19 (8.5 to 43.3 %) without
- Leaves the human corridor: 6 of 19 (15.4 to 54.0 %) samples with route text, 9 of 19 (27.3 to 68.3 %) without
- “Right lane” in its reasoning: 9 of 120 (4.0 to 13.6 %) strings with route text, 6 of 120 (2.3 to 10.5 %) without, matching our rule; on multi-lane UK roads it can be correct
- Parked cars in its reasoning: 15 of 120 (7.7 to 19.6 %) strings with route text, 13 of 120 (6.4 to 17.7 %) without, matching our rule
- The model saw views reprojected from our 360 camera, not the cameras it was validated on: a bench result, not its behaviour on the road.
Why not the Act's test: “careful and competent” is the legal test in the Automated Vehicles Act 2024, section 2(2)(a), rooted in the Road Traffic Act 1988 s.3ZA; this page measures an observed envelope from one driver and makes no claim under that test.
“authorised automated vehicles will achieve a level of safety equivalent to, or higher than, that of careful and competent human drivers.”
Not measured here: headway, gap acceptance, time of day, roundabouts, jerk. Roads off the measured loop are drawn flat and dark: not measured, and nothing here reads as safe. Method and sources
Map data © OpenStreetMap contributors (ODbL). Contains Environment Agency information © Environment Agency and database right, OGL v3.0. LiDAR and trajectories by MapMap.