It may come as something of a shock to many here, but Google have done extensive research (to add to the already extensive research done on aircraft pilots) and have concluded that when humans are involved with operating complex heavy machinery, there are really only two varieties of automation that are useful.
Firstly, there is the technology that is being steadily fitted to all cars now, which is stuff which improves how a human controls a car. This is stuff like blind spot warnings, braking assistance systems, anti-collision radar and various lane-keeping aids, together with ABS braking, improved suspension and so on. All of this requires the human to do the driving, and the tech just tries to help the human out whenever it can.
Secondly, there is fully autonomous driving. This is where the car and its systems are doing all the driving work, and the human merely has a big red button to hit in times of panicked emergency, together with some sort of very low-speed movement control for parking the vehicle somewhere that the autonomous systems cannot.
Tesla seem to be trying to extend the assistance technology into the driverless technology niche. The problem here is that humans are really bad at not driving but remaining alert; not just slightly bad but truly, dangerously terrible at this to the extent that this system is actually more dangerous than a person driving an onmodified, unaided car by themselves. Until Tesla realise or admit that their systems don't sit in either of the sweet spots for this sort of technology, they are going to carry on having problems.