Any victim should be dealing with.
Either a car is fully autonomous and carries the same liability and associated penalties as a car driven by a human or it might as well not exist.
Doing this half-assed is worse than doing nothing in my opinion, drivers that get to point at their car and say 'it wasn't me' is not an option.
Let's say for sake of simplicity that the system can reduce the risk with 50%. Under normal driving conditions when the risk of hitting somebody is say 0.0001%, you instead get 0.00005%, that can't be bad. If you provoke the system by actively aiming at a human you by definition have 100% risk of collision without the system, with it you have 50% of 100%=50%, I wouldn't want to be that pedestrian, even if the system could reduce down to 1%.
Autonomous drive on the other hand, you give the car control, as opposed to the car taking it from you (because you have been a bad driver). If you give the car control then you should know the systems limitations and which responsibilities you are giving away. What is control anyway? Is cruise control a half-assed version of autonomous drive or not?
In these cases the the vehicle and to some extent the environment is much more controlled and it configured for optimal operation.
On the consumer side it's a completely different story because not only does the autonomous driver needs to outperform the human driver it needs to do so in a case when you have compound errors if the driver of the car does not trust what the auto pilot is doing you'll get into more accidents not less, and in a market environment where you have a very broad range of features you can inconsistencies which lead to more accidents.
And this is where you hit the UX issue, all autonomous drivers have to present a reliable and consistent user experience to avoid compound errors that arise for the involvement of the human (jeez that sounds eerie too natural these days) driver.
When I step into a car with an auto pilot I need to know that it will get me from A to B with minimal supervision and will on it self avoid obstacles no matter what car I step into.
If car maker A is allowed to sell it's auto pilot without collision avoidance for example while car maker B has it it will create confusion which leads to accidents, if car maker C then decides to sell an auto pilot feature which is nothing more than a glorified cruise control it can lead to even more accidents. This is why some standardization is really needed with this tech getting off the ground.
The standardization also has to go beyond sheer features it has to provide a consistent user experience across the board so things like acceleration, breaking distance, driver alerts etc. need to be more or less consistent across different cars from different manufacturers or again you'll can get an increase in accidents due to driver errors.
"Human killed by robot car" will definitely be lapped up and even if statistically it may be better than "Human killed by human driver" the one is the exception, the other the rule and so the one will get airtime.
Marketing this kind of thing properly is going to be extremely hard once the casualties start mounting (and they will, no matter how good the software is).
But that's still a compelling number. Just as most people thought malaria was some quaint disease that inconvenienced the British Empire until Bill Gates got on it, I think road deaths are potentially at least as compelling a target, especially once people look outside Europe and North America.
I don't think that's even remotely accurate.
I see Autonomous driving like iOS vs Android when you buy an Apple device you know what to expect (or atleast used to before they decided to fragment their SKU's to hell) each device comes with the same features and the software works exactly the same. The Android ecosystem on the other hand is fragmented to hell and each devices comes with completely different features, software and user experience.
This can easily kill autonomous driving because it unlike poor app performance self driving cars can actually kill people if the self driving experience I get on a BMW is drastically different than what i on a Honda or a Volvo I won't be confident in using any of those, not to mention that at the moment you can't even get reliable experience on the same make and model and it's a feature not a bug in most cases because car manufacturers want to monetize self driving so for every (even basic) feature you'll have to pay extra.
Self driving cars need to get a standardization and they need it fast a fixed set of features that all cars have to properly support to be able to market their self driving cars in the US and Europe and this has to be a dichotic standard you either support everything in that standard to a satisfactory level or you don't get to turn any of those features on.
The last god damn thing I want to worry about is if my car's pedestrian avoidance app was reactivated or not after the weekly firmware update, but sadly enough i foresee some one getting run over because the driver of a "smart" car did not accept the new EULA that was pushed over the weekend before the industry gets any shred of sense.