I work for an autonomous vehicle company and can tell you it's not nearly as close as people think. Sure, cars can do some very impressive things these days but a completely self-driving car is a long ways off and most people believe Level 3 is impractical. There are plenty of studies that show people cannot simply regain attention quick enough to take over when L3 fails. It's far more practical to have a L3 human car. I..e human does the work and car jumps in when you are about to depart the lane, rear-end someone, etc.
There are a couple of fundamental issues:
1) AI (actually machine learning is the right phrase) is a great tool but it can only solve part of the problem. last month Tesla's were confusing the moon for yellow lights. They use AI/ML for that which requires the system to be fed millions of images. Need to fix the moon/yellow light issue? Feed more images. Problem is with every edge case you need to feed it more data so it becomes asymptotically impossible.
2) Humans are actually very good drivers. A lot of talk is around 'self-driving will be safer'. Sure, some day it may be, but right now it's not even close. In the US there is only 1 fatality per 100 million miles driven. And those miles include driving in Minneapolis in January in a snowstorm. So, the bar to be 'better than a human' is really, really high.
3) There is no business model right now demanding these. Toyota now offers ADAS systems (LDW, FCW, etc.) for free. Sure, that's built into the base price of the car, but the reason they include it is because no one pays for those when it's an option. Uber wanted robot-taxis to save money but their effort (called ATG) eventually failed. It was given to another autonomy company for free along with a $400M incentive to take it off their hands. Now it's trucks. But the truck demoes have all been limited to straight highways in places like Texas. Then when they get close to a city a driver hops in and finishes the route. They have zero ability to navigate in congested areas. Sure, it can drive highway but this means that you'd have to add a depot on top of the existing depots. Not exactly cost effective.
4) The people in charge now this is true but there is so much money in the space now they just keep powering ahead. There was an interview recently where a self-drive company CEO was quoted as saying, "this is a generational problem. My kids will be working on it and it may never be solved." Of course, that video has since been buried because once these companies got their SPAC $ they need to justify their bloated valuations.
Software may be more attentive but it's still fundamentally stupid at handling anything unexpected. So, the best approach is attentive human drivers that are augmented with a supervisory autonomy to help when there is lack of attention, not the other way around.