If FSD is the goal, then why not fix the simple things now?

[plumazul]

I just returned from a 2,500 mile road trip in my Model S. I tried to use NOA but I found it exhausting and often turned it off in favor of vanilla AP. NOA requires so much supervision. For example, it tries to pass on the right which I never do (should be a setting) and it even tried to change to a lane that was ending 50 yards down the road or into a lane that obviously had slow moving trucks a short distance ahead! But these things are probably difficult to fix and so I will wait, but what I don't understand is why two standing issues with AP have not yet been fixed.

1) Why insist on driving dead center in a lane?? This is not how humans drive. When driving with two lanes in your direction with a standard shoulder, almost all drivers, (and yes, I was watching) bias to the right in the right lane and bias to the left in the left lane. This bias is exaggerated when passing/ being passed by large trucks, and is REQUIRED when passing over-sized loads. In every case, AP insist on dead center. The result of this behavior is constant disengagement.

2) Why does AP again have problems with "fat" lanes? This was fixed (for me at least) back in 2018 and has now returned worse then ever. To me, the solution to this is simple. Compute a virtual lane based on the center stripe and pay much less attention to the white edge stripe, if it is there at all. I have a 5 mile "test course" near my home that I use to test each new release. In the last two Qs of 2018 AP was perfect on this course. Now it aborts in 3 different places and drifts to the center of the lane in many other places putting me at risk of hitting parked cars.

I am slowly losing hope.

 

[electric_avenue]

Completely agree - lane centering is a dumb idea - they should anchor to the lane markings on the driver side - lane centering creates too many dangerous situations with erratic behavior.

 

[Rockster]

My 2015 AP1 P85 handles fat lanes flawlessly. My Model 3, not so much.

I've found that turning down the NoA settings helps reduce its frenetic lane changing. I tried Mad Max but that was too absurdly agrressive, changing lanes every 20 or 30 seconds. Turning it down one level helped, but not enough. One below that, though, works much better. I find myself gesturing to manually change lanes more than I did, but the number of times I need to override the lane change is much less now.

It does seem that Tesla has extremely optimistic goals if they think the software is going to be effectively FSD by the end of the year. They must be sitting on some seriously improved code for that to be true.