Interesting article too -)
Thinking of forensic evidence out of an AVL5 crash lets think about where are the important areas or layers of 'gold first'?
Postcrash a vehicle often moved away from the location of crash by forwarding energies. So there is a Crash Location CL and a Real Location RL. Actually by 3D-Laser Scanning coupled with GNSS+ we reverse simulate from RL to CL. But there is a general problem invisible all the time. At the crash timestamp a specific constellation of external factors influencing more or less a crash is no more available. Cars continue to drive further, weather- traffic conditions change constantly and mobile traffic users like bicyles, pedestrians and others are no more available for evidence.
For this we split vehicle sensors in eyes and sensors. Eyes have seen things around visually but sensors only by detecting a bounding surface of objects. All in-car data is fed into the zFAS (A8 AI) as the car mainboard. There are at least 4 major CPUs or GPUs aggregating the sensing into mapping and policy making.
But which data remains in-memory in-car for the RL or was moved to the CarCloud of the manufacturer? I guess more and more data goes to the BigData hungry CarCloud and is lost.
For now we are testing OBDII dataloggers (weDIY) to wireshark the car LTE modem.
Only to prod with a reality check (if needed)
https://
jaclaz