as I have to drive today anyway, I will immediately get a chance to test. Looking forward to checking my UPL. I do not have high hopes.
We are all expecting another honest assessment you lucky bastard!
So, not too much time yet, and I did take videos, but no time to process them for now (possibly ever).
Quick impressions:
My
UPL: failed twice, was successful the third time.
- It is still turning the wheel to the left before it gets to the turn when creep network is inactive.
- when the creep network turns on, the wheel turns back to the right (new!), but this takes a while and still leaves the car in the wrong place (too far to the left, cutting off left-turning traffic). Right now: It’s a two-stage (maybe 3?) process: 1) proceed to stop, then 2) wait for creep network to turn on and 3) start creeping.
- the first “failure,” it made the turn, but I don’t like the behavior where it goes with oncoming traffic on the far side. My turn is the type where you just wait for both directions to clear and go. If it had gone aggressively, it would have had plenty of room, but it did not of course. It needs to be much more assertive if there is not a lot of time. No one should be slowing down for me! There was a car coming on the far side, and close enough that they changed lanes to avoid me when they saw me go (I disengaged and sped up so they never even came close to passing me). It seemed like the car was planning to wait in the median merging lane, which is completely incorrect. You use that lane to get up to ramming speed!
- second failure, it creeped way too far! I disengaged because I wasn’t sure it was going to stop and there was traffic coming and I did not want to freak them out. It isn’t fast enough to creep (too much delay), then when it creeps, it creeps too fast!
- I do not know if Tesla will ever implement the behavior where it ignores the stop line (which is usually way too far back for visibility - though it does not show in Street View, there is a stop line painted now on my turn (no yellow line in the middle though)), and just proceeds to the correct spot. But if they do not (might be illegal, even though nearly everyone does this single stop rather than double stop), they need to make the whole process much faster (stop at line quickly, then without delay, start creeping slowly).
- in any case it sticks out way further than needed. My turn has very good visibility; though creeping is needed past the stop line, there is not a need to proceed as far as it does. Reported.
- on the successful attempt, there was no traffic. It used the merging lane in the smoothest fashion I have ever seen. Nothing to write home about, but progress! (It’s also been wildly inconsistent on this in the past, so curious how it will do on a few more tries; could have been luck.)
Other observations:
- it failed to enter a two left-turn lane area in time (so it crossed late, across the solid white line). There was no traffic. I think this is a regression, probably due to the deep lane guidance stuff. It was signaling in time, it just did not move into the
area. Reported.
- The torque sensing seems much more touchy. I have rarely seen the blue screen nag, but this time on the freeway I disengaged AP twice, attempting to clear the torque nag (a standard tug (which of course I was already applying) did not work). This also happened on a surface street and I just disengaged there since it would not “take” the input.
My guess is this is a temporary tuning issue to keep people extra alert, that will be addressed. Or they dialed it up on my car only, just to spite me for all my b**ching about the strike reset. Reported.
My prediction: a bunch of new strikes for people who just had their strikes reset. No more resets please! There is no excuse, even for hardware failures. If it’s not working turn off FSD/AP. That is my line in the sand.
- Overall, on the few turns I did take, the steering was pretty good. On one of them, the unwind was a bit jerky. I would say overall improved smoothness, with a lot of work to do. I don’t think I have enough seat time yet to fully evaluate this.
- It’s very slow and falls behind other traffic and does not get up to the set speed in a timely manner without accelerator input. No changes noted here.
- On the slowing for traffic lights (no testing of stop signs), it was generally smoother than before. Still definitely not butter smooth. Need more checking to evaluate.
- My
exit still results in the car zooming into it at 75mph. It jams on the brakes (actual brakes) once it crests the blind hill, and slows down on the downhill. This is not new behavior though. Really bad. I have reported this in the past. I would drive this way if I were trying to set a speed record, except I would brake later in that case and use all of my tire traction.
The general environment here is well marked arterial surface streets, similar to Spring Canyon (plus a long section of freeway not of relevance except for the torque nag). Occasional turns at lights but mostly just driving straight.
Pretty much what I expected, except the creep behavior is not as good as I expected. Looks like they need to tune that a bit still, to get the right behavior. This error has been widely reported in many of the above posted videos, so it generally seems like a well-known issue. Hopefully 10.69.2 will adjust it a bit!
Definitely seems like there are improvements in several key areas, but I need to drive with it more to be able to assess how much (or even whether they are actually improved!) exactly.