As long as unexpected events can take place, you can't expect the Tesla to react properly in all cases. If the Tesla can't do a temperature measurement how would it know there's a fire in front of it? In one of the first deadly accidents the Tesla identified a truck as a "sign" and plowed into it. That's because the ride of the truck was higher than expected. That's the key word - "expected". And how are drivers to know what the Tesla is expecting or not expecting? I think we are way too early into AI development to allow it to control things that can kill us without complete human supervision. Understanding the fatal Tesla accident on Autopilot and the NHTSA probe [electrek.co]

I saw a video just the other day that showed the "driver" of a self-driving car sound asleep behind the wheel, going down the freeway near LA.

A backup plan such as taking control of the wheel/system override wouldn't be a bad idea. The information centers in vehicles might be more advanced in the future, information on destination updated in real time.