Why would all self driving cars not be in constant communication with all others within their stopping distance using a standardised protocol? It would seem a trivial thing to implement in the scope of self driving.
What sort of range would this "trivial" communication have? Oncoming traffic will be closing on you at your combined speeds, so probably 80mph and up in a number of situations. What sort of calculation delay would there be if two cars negotiate a mutual action? Regardless of whether they were in communication before, this would have to be done after the emergency occurs and it wouldn't be a trivial delay.
If both cars brake and lane share there may not need to be an accident. Certainly there is no risk if your car knows having 'spoken' to the other car, that it will stop and in what distance, and your car knows it too can stop in the remaing gap.
If.If.If. My point of view is that autocars can talk amongst themselves all they like and it will likely come in helpful to crowdsource knowledge of traffic jams and roadworks, but a car should never put itself into a high-risk position (and it's certainly not 'no risk' as you say) based on what another cars 'plan' is. To do that is to assume that the other party has all the facts and then bet a variable number of lives on it.
it won't think, it will know for a fact because the other car will have confirmed it.
I will accept that autocars know something 'for a fact' when they have physical evidence supporting it, e.g. the cameras show the other car stopping. Anything less/earlier isn't good enough to bet lives on.
I realise this comes across quite negative about autocars, but this couldn't be further from the truth. I would love to see them come about asap and agree that they will improve safety for all massively, I just don't think split second coordinated responses are feasible; it throws a lot of complexity into a time critical situation.