It's all a little pedantic
The idea that something needs to be correctly identified down to the nth degree is a little pedantic. The main rules will be "if there is an object in the road, don't hit it".
Whether or not it can see an object in the distance will be down to it's sensor's range and the time it would take the car to stop in those conditions. It needs to work out how far it can see. If it knows that it can only see 30 metres in front (for certain) then it needs to go at a speed which will make it sure that not only can it stop if needed but that it can stop gently, that limits it's speed, down to perhaps 7 mph. It doesn't need to know the top speed limits written on a sign.
On top of what it can see ahead of it, it also needs to take into account the surprise object suddenly appearing from the left from a crowd of moving objects (called people). It doesn't need to identify a child from an adult, male from female, just that it is an object.
On many roads, those moving objects on the path are only 4 ft away from the car. If it takes into account that any one of these objects could suddenly come from the side, and fall in front of it, then just knowing that this is a possibility will keep the car going very slowly just in case.
If the AI car is made safe, it is also made useless, since it moves too slow.
If the car is made useful, it is now going too fast to be safe, risks have been accepted, and so little details like correct identification are now less important.
For me, the question is, do you want to accept a car driven by a human making a mistake or an AI car. The human can be blamed, pointed at and taken to court.
The AI is a faceless corporation backed by government and the legal system that allowed it to be on the road. When it makes a mistake, who are you going to blame, and more importantly how are you going to get "justice" for the dead child? Good luck with prosecuting government / corporation / legal system once they have accepted that the risks are ok.
Leave humans in charge of a car. Just use AI to help avoid human mistake and enforce "a safe speed for the conditions"