Test driving cannot prove safety.
Automated driving fails on the rare mile. No fleet can drive enough of them, so the evidence has to come before the road.
OutRes (Outlier Resilience LLC) is an early-stage applied research venture in Michigan, working with teams who must validate a driving stack they did not write.
01The arithmetic doesn't work
Human drivers average 1.09 fatalities per 100 million miles. RAND calculated what proving an automated vehicle beats that would take.
“It is not possible to test-drive autonomous vehicles to demonstrate their safety to any plausible standard, even if we assume perfect performance.”
Kalra & Paddock, Driving to Safety, RAND, 2016
ISO 21448 names the class of risk: harm with every component working as designed. It asks you to shrink the set of unknown-unsafe scenarios. It does not say how to know how much is left.
02The tail is not exotic. It's Tuesday in February.
The conditions that degrade a perception stack are mundane, frequent and regional. A test program assembled somewhere sunny will under-weight them. A model fitted to clear conditions meets one it wasn't trained on, and the failure is quiet.
- Degraded optics
- Low sun, wet glare, spray, a salted lens. The image still arrives, degraded.
- Vanishing road structure
- Markings under snow or wear. Models that lean on paint inherit a dependency the road doesn't keep.
- Compounding conditions
- Dusk, precipitation and a worn surface at once. One-factor-at-a-time matrices miss the combinations.
03Research
Claims here arrive as papers with methods attached, and with the code that produced them.
04Compare notes
If you validate an automated driving stack, your own or a supplier's, we'd like to hear where your process runs out of evidence. We're not selling anything yet.