Not even close. 35K annual deaths and several million injuries is in no way remotely acceptable. Imagine if air or train travel were this deadly. The NTSB would shut it down immediately until a way could be found to make it much safer.
Also keep in the mind the "monitoring" systems you decry have been incorporated into modern aircraft flown by highly trained pilots. Putting similar systems in automobiles makes them safer. All these systems do is reject nonsensical inputs which always have bad outcomes. For example, if you start drifting off the road the system won't let you leave the road and crash, or in city driving leave the road and hit people on the sidewalk. What's wrong with that? Or they won't let you follow at less than a safe distance. In urban areas they shouldn't allow you to exceed the posted speed limit or accelerate at more than gentle rates (i.e. no more "stop light grand prix"). Basically, they won't kick in if the vehicle is being operated safely and competently. They exist for those moments where you might have a lapse of attention, or just want to drive recklessly. The end result of this is saving lots of lives until fully automated driving is ready for prime time.
The rejection of "nonsensical inputs" makes significantly more sense when there are fewer options in general for maintaining a stable flight pattern. As to your example of bad input =100%= bad output, you are terribly incorrect. There are a vast number of reasons an attentive driver may purposefully swerve or edge out of their appropriate lane of traffic, even up onto a curb in urban environments. The roads are littered with debris, roadkill, people, animals, etc the list goes on. A safety system that actively prevents the driver from being able to respond appropriately to changing road conditions is no safety system at all. As is the complaint leveled towards the driver, "they work until they don't." As it stands now, the ultimate liability falls on the licensed person behind the wheel. Refer back to my apparently nonsensical claim that perhaps licensing standards should be raised significantly (as long as we're comparing to pilots this doesn't seem nonsensical).
As far as systems that restrict acceleration, while I can see the argument in an urban environment, I have very practical experience with governed vehicle speed/acceleration due to a previous employer. And there is nothing more unsettling than being fully aware, as you are "the guy" holding up significant volumes of tractor trailer traffic, and that your vehicle will not respond to your inputs if you need to move the hell out of your lane, or accelerate momentarily to avoid collision, and cut power to limit your speed when you're in the far-right lane with two lanes filled with tractors to your left, no shoulder to the right and a downhill stretch that every tractor behind you fully intends to take advantage of for building steam to make it up the next rise, whether you're in the lane or not.
I have stated it elsewhere, I'll repeat it here. Self driving cars are a pipedream, despite all the countless engineering working towards the goal. The first time a mercedes with an octagenarian at the wheel opts to take out a crowd of schoolchildren to save the driver, the first time people realize that their fancy new bathtub refuses to let them leave the driveway on an icey side street losing them their job, the factories will be burned down.
Back to the adaptive headlight issue, and in anticipation of yet more moderator redactions... The camera based system for the automatic high/low beam selection on my cx30 seems to have some strange issues with retroreflective signage at times, where it seems to want to give other drivers seizures. This same camera system controls the intermittent wipers, which keeps the camera lens fairly clear, but also causes some annoyances with very irregular wiper action. Quite distracting. I have not had sleet/snow to see how these systems respond, but my current assumption is "not well." If people pay a premium for these adb features, especially if it is on a rental basis which so many vehicle manufacturers seem to be heading towards, and they crap out in the first inclement weather the driver experiences, reverting back to the current "loose screw between the seat and wheel" control, people are also going to be irate. I can't imagine a solely camera based (*cough* tesla *cough*) approach being effective at modeling the changes needed in conditions of poor visibility. Perhaps it would be more effective to tell everyone to just forget their lowbeams exist at night, if the goal is to maximize the ability to see.