Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Poll : Which is better - 10.4 vs 10.3.1

How would you compare your experience with 10.4 to 10.3.1 ?


  • Total voters
    80
  • Poll closed .
This site may earn commission on affiliate links.
I went with 10.3.1..

some of my problem areas are WORSE, some of the okay things - like smooth driving just on a plain straight street, are worse. The car is doing slight ping ponging between the lane markings as if it is over thinking it, or over interpreting a slight curve of the lines or vehicles or other objects around. It’s almost MORE uncertain than it was in 10.3.1
 
10.4 slightly better for me, as phantom regen braking has been reduced.
For me it:
  1. still stops at empty round-about yield signs,
  2. still stop well before stop signs,
  3. still prefers the the center of unmarked two way roads even with oncoming traffic,
  4. still make a right turn at intersection into the parking lane,
  5. still signals turn at curves in the road,
  6. still can find my street to turn into half the time,
  7. etc., etc..​
I'm sure the FSD programmers are working very hard, but at this rate of noticeable improvement, I think Waymo and Cruise may be offering geo-fenced autonomous taxis in a dozen major cities by the time FSD can drive as well.
 
  • Like
Reactions: S4WRXTTCS
One thing I really don't like on both 10.3.1 and 10.4: This has happened twice, once on each. It never happened before on AP (>15 months).

On freeways when FSDbeta sees a car merging on the right it'll brake (sometimes hard) to seemingly match position and speed (!!). Normally I'll keep my speed up or accelerate to get out of the way and avoid getting sideswiped. I never brake and put myself into a hazardous position.
 
definitely two stacks
Old Stack >65 limit
New Stack <60 limit
Real PITA when it flips back to new stack because it re-enables the auto high beams etc
Even though its still the old stack, the flip to vision from radar makes it much less sure of itself on some roads. Too many hesitations to be comfortable.
I always felt the radar releases much more confident than vision - apart from the sensitivity to bridges and dark shadows :D

The radar version is vastly better in rainy conditions.

For the first time in my entire Tesla ownership (since 2015) the adaptive cruise only had to slow down because of weather, and it wasn't even that bad.

I went from having probably the best SW build of AP/TACC before the Beta (during the safety score) to having something atrocious. Way more phantom braking while on the freeway, way more weather related issues, etc.

I'm curious to see when Tesla will release Vision only to the entire fleet because that if its a long time I might have to exit the Beta just to restore radar for a bit longer.
 
10.4 slightly better for me, as phantom regen braking has been reduced.
For me it:
  1. still stops at empty round-about yield signs,
  2. still stop well before stop signs,
  3. still prefers the the center of unmarked two way roads even with oncoming traffic,
  4. still make a right turn at intersection into the parking lane,
  5. still signals turn at curves in the road,
  6. still can find my street to turn into half the time,
  7. etc., etc..​
I'm sure the FSD programmers are working very hard, but at this rate of noticeable improvement, I think Waymo and Cruise may be offering geo-fenced autonomous taxis in a dozen major cities by the time FSD can drive as well.

5 is why I tend to not use it in those situations when I have someone behind me
 
For me 10.4 is vastly smoother, which I really appreciate. Acceleration ,deceleration, and turning all seem improved in that respect. It does seem like I’m seeing more weird planning behavior than I did on 10.3, though, like turning right when it’s supposed to turn left, or suddenly jerking towards one side of the lane. Sometimes those behaviors are alarming.
 
  • Like
Reactions: powertoold
Update:

This also happens on surface streets. FSDbeta sees a car merging on the right and brakes to seeming maximize chances of getting sideswiped. I was 1 or 2 cars ahead of someone merging and FSDbeta tried to put me right beside them.

Doing nothing (maintain speed) or speeding-up (avoid) would minimize probability of getting sideswiped.

Braking accomplishes the following--
1) Increases probably of getting side-swiped and 2) increases probabilty of getting rear-ended.

One thing I really don't like on both 10.3.1 and 10.4: This has happened twice, once on each. It never happened before on AP (>15 months).

On freeways when FSDbeta sees a car merging on the right it'll brake (sometimes hard) to seemingly match position and speed (!!). Normally I'll keep my speed up or accelerate to get out of the way and avoid getting sideswiped. I never brake and put myself into a hazardous position.
 
Agree
Reddit demographics is likely younger & less patient.
Agree. Today we're seeing Electrek reporting on "a serious phantom braking problem" so the overall situation I believe is getting worse with 10.4.
Tesla needs to respond quickly as external complaints continue to increase and with it their exposure to regulators. Lets hope 10.5 begins to address the problem.
 
Agree. Today we're seeing Electrek reporting on "a serious phantom braking problem" so the overall situation I believe is getting worse with 10.4.
Tesla needs to respond quickly as external complaints continue to increase and with it their exposure to regulators. Lets hope 10.5 begins to address the problem.

Unfortunately I think phantom braking is a tough issue to crack and will only be solved slowly.

Basically its a balance between false positives (phantom braking) and false negatives (crashing into objects). They just need to keep adding more training data and optimizing ...

Voxel mapping using videos should help greatly. But that might be a few months away.

Overall, I expect FSD Beta to be in good shape for general rollout only by the end of 2024.
 
balance between false positives (phantom braking) and false negatives (crashing into objects)
I was demoing FSD to someone interested in a Tesla yesterday and had a braking incident that would only happen with FSD thus resulted in some explanation of what not-FSD Autopilot would have done:

Someone was speeding out of a parking lot on the left side of the road and ended up turning right. FSD Beta incorrectly believed the vehicle was going to continue straight crossing our path whereas not-FSD would have only braked once a vehicle was directly ahead. Unclear if it's actually a false positive for FSD and false negative for not-FSD as maybe this is the "correct" safe driving behavior? Most people would assume this fast moving vehicle wouldn't actually pull out into heavy traffic, but I suppose it's possible.
 
wonder what's the average age here vs reddit.

According to a survey of U.S. adults conducted at the beginning of 2021, 36 percent of respondents who indicated using Reddit were aged between 18 and 29 years old. Of course we don't have the age information for this forum but would think it would be older on average. But to @EVNow's comment that Reddit users would be less patient I would agree. The overall point I'm trying to make is the amount of highly critical feedback on FSD has grown significantly in the past month directly related to the FSD beta testers rollout. It has become a double edge sword to Elon and Tesla.

U.S. Reddit reach by age group 2021 | Statista
 
  • Informative
  • Like
Reactions: nvx1977 and EVNow