Smoothing algorithms seem to be non-existent or written by a first grader who doesn't even know how to code. No normal human being changes lanes that quickly and the Tesla is dangerous to do it like that. I really don't understand how they could have such crappy software. And just for the record I have been a software engineer for over 30 years. I have had several false positives under bridges and overpasses where the car brakes hard enough to easily cause a rear end collision. On one freeway at night I was in the right hand lane of the freeway and within a 3-minute period, it tried to jerk the car off the freeway at two different exits. And when I say jerk I mean it really pulled hard and quickly to the right. This is downright dangerous and I thought they had solved this problem a long time ago. The lane markings were clear at the time. All of this AP 2.0 experience was in the model 3 by the way which I just sold. I also have a Model S and the algorithms for that old autopilot (AP 1) software are far superior and are very smooth and predictable for the most part. I'm kind of pissed off that Tesla decided to try to do everything in house and that because of some stupid spat with a company we are all suffering with an absolutely abysmal and shity version of autopilot. They should have just worked out whatever problems they had with mobileye and kept going with them. The problems that they are trying to solve must be extremely complex and multifaceted and to think that you are just going to hire a bunch of people and code that *sugar* quickly is ridiculous.
At the very least Tesla should be monitoring all breakaways from auto-steer and then replay the scenario back to themselves so they can see why someone decided that they didn't want to go flying off the freeway in a violent maneuver that the autopilot decided it was time to do RIGHT NOW!!!