i assume the decisiontree for the algorithm branched off like:
Quote:
if bicycle is not at xwalk;
then rider is travelling parallel (not perpendicular);
> goto drivealongtraffic;
...

people smarter than me should calculate the amount of auto fatalites in this country proportional to how many total miles driven for the previous calendar year using the maths necessary (stochastic processes, guassian distributions, probability density functions, calculas, fourier transforms, ...).
then we would prove that robot drivers are statistically less safe than humans.