Nice work! Some of the situations in the drawings imply that the car needs human-level commonsense to perform properly. For certain things like 'fire in a tunnel' or 'tornado crossing the road', there could be sufficient time to warn humans to take control or at least give verbal instructions, while the car is slowing down drastically or even turning away as a safety-first precaution.

The challenge for the autonomous car would be "how to know that it doesn't really know". I wonder if some existing philosophy or theories are applicable for such a purpose in the real world. Does anyone have pointers?