Model S P100D AP2 driver here and agree with everything
u00mem9 posted in September, especially Tesla returning to Mobileye technology via Intel to limit the volume of egg on their faces. AP 2.0 or 2.5 makes no difference and I state that because the problem limiting the function of Tesla autopilot now is not hardware, it is software.
I spent the past 10 days driving an AP 1.0 MS P90D loaner while my P100D was at the SC for repairs covered under warranty and I can say that the AP 1.0 drove with great confidence on the highway. It tracks vehicles in 3 lanes (the lane you are in, plus one on either side), and thus can preemptively and gradually slow your vehicle "like butter" before a vehicle from one of the other 2 lanes switches into your lane. It can tell the difference between a car, truck, or motorcycle, and can therefore predict behavior of those types of vehicles more accurately. It's perception of lane lines is dead accurate and generally the vehicle says perfectly centered. It can accurately stop your vehicle based on the abrupt braking of the car 2 vehicles ahead of you (I have experienced this twice in the AP 1.0 car, but not yet in the AP 2.0 car, not to say 2.0 is not capable of this, I just personally have not experienced it). Its automatic high beam function is far more facile than the AP 2.0 vehicles. And it has functional automatic wiper which the 2.0 vehicles do not.
After returning to my P100D, I experienced the complete opposite of the above features I described above. The car drove so erratically due to inability to perceive lane lines correctly that I stopped using AP entirely.
This lack of functionality has nothing to do with hardware, as my P100D has 7 cameras dedicated to AP (AP 1.0 has only 1), and more powerful radar and ultrasonic sensors than 1.0. Short of adding in Lidar (which could happen with AP 3) better hardware means nothing. The difference in functionality has everything to do with software. The AP 1.0 car has the Mobileye system on chip IN ADDITION to Tesla's proprietary ADAS autopilot software, whereas the AP 2.0 and 2.5 cars only have the Tesla ADAS. The Mobileye chip encodes the deep neural net functions essential for autopilot. These deep neural net functions allow for visual object recognition, which range from accurate perception of lane lines, to the physical characteristics differentiating cars, trucks, and motorcycles, to finer points such as road surface conditions.
Mobileye spent over a decade developing this proprietary technology, and this technology alone. When Mobileye and Tesla split, Tesla thought they could develop the same technology on their own and get by without the Mobileye chip. No doubt they could have, but unfortunately they did not have the talent to recapitulate a decade worth of deep neural net software in 6-12 months as they tried to do, particularly given the thousands of other tasks they have to perform to succeed in the complex automotive industry. This is why a return to system on chip (SoC) deep neural net software is nearly guaranteed.
I predict that Tesla, being a good company, will eventually install the SoC encoding the deep neural net functions on the AP 2.0 and 2.5 cars once it decides which vendor to go with (be it Intel/Mobileye or another vendor). Until then, the AP 1.0 cars offer a solid level 2 autonomous driving experience on the highway, whereas the AP 2.0 and 2.5 cars offer a less predictable level 1-2 experience.
On the other hand... the actual experience of DRIVING the Model S P100D is unlike anything else in the world!!! It is a far more fun, responsive, and exhilarating car to drive than even the P90DL due to better handling, faster acceleration, and better traction control as well as longer range.