Poll : Which is better - 10.5 vs 10.4

[EVNow]

Note : Public poll.

Consider how the two releases performed on your routes. Which release did you like better ?

This was the result of the last poll.

Poll : Which is better - 10.4 vs 10.3.1

Note : Public poll. Consider how the two releases performed on your routes. Which release did you like better ? This was the result of the last poll.

teslamotorsclub.com

 

[spsheridan]

Is it possible the last item in your poll was intended to be "10.4 was much better" rather than "10.5 was much better"?

 

[EVNow]

Is it possible the last item in your poll was intended to be "10.4 was much better" rather than "10.5 was much better"?

Oh yes. Asked mods to change it.

 

[EVNow]

Here is what I just posed in the other thread.

Had a comparatively bad drive in the morning dropping off kids.​

​

Its the same route I've done 20x over the last month with FSD - there is some variability, even with same release. So, have to wait for a few more drives for a reliable report.​

​

- On the first right turn from stop, suddenly jerked the wheel to the left - so got disengaged. While I've written a lot about the jerkiness when turning right from a stop, none of the wheel action has been this much.​

- At an intersection where the side roads have stop signs, the shoulder starts out very broad and then narrows. On rainy nights, the car had got confused once and gotten into the shoulder - but today in broad day light on a dray day, the car decided to use the shoulder rather than the lane it has taken every day. Second disengagement.​

- At a signal when going straight, the car decided to go only 100 ft or so and then stop. Not sure what happened - had to take over. Unfortunately hadn't turned on the camera. So, third disengagement.​

​

Apart from the above 3, had some usual disengagements.​

​

- At a traffic signal where the car needs to turn right, it is supposed to be no-turn on red (the signal says "right turn signal"). But most people turn and honk if someone doesn't. Had to disengage here because, the car stopped well short of the intersection and for me to even press the go pedal, I need the car to creep. Not sure we can control the creep by pushing the go pedal, without the car deciding to take off at the wrong time.​

- Roundabout that is not in the map​

​

Other observations ..​

- The car seemed to keep more to the right on unmarked roads. Could be just the normal variability.​

- When finishing the drive at home, the car came closer to the curb. New behavior or just by chance ?​

- Turns when already moving have always been variable. Some smooth, others jerky. Today all the turns seemed jerky.​

​

So, on the basis of this one drive, I've to say 10.5 was worse than 10.4. But need more observations to confirm.​

 

[hellocar]

10.5 seems steadier for me, positioning better in lanes more often, deinitely fewear weird planning errors.

 

[TheBrumBrum]

10.5 indeed feels better with lanes and unmarked multi lane intersections. That's crucial.

However it feels jerkier to me, leaves much more space than I appreciate. It also moves much slower at lights than 10.4 and I'm now getting a lot of phantom breaking which was no issue before. It's usually not much of a speed change even.. 1-2 mph. Might be related to the time of day I drive and the sun just being at a bad angle for a few weeks? Guess we'll see.

 

[nvx1977]

I voted 10.5 being much better than 10.4, but that's mainly because I had a very poor experience with 10.4, which makes it feel like a wider gap. If it were between 10.5 and 10.3, I likely would have voted that 10.5 is slightly better.

Things I've observed that perform more consistently:

- less jerk/jitter with the steering wheel for turns. It's actually kinda black/white: either it turns very smoothly, or it's really erratic. I haven't encountered something in between. But the vast majority of my turns have been surprisingly smooth. 10.4 trained me to be hyper-vigilant for turns, and it's been nice to be pleasantly surprised turn after turn.

- car stays more to the right on unmarked roads. I used to just turn off FSD whenever on these roads because of the risk of head-on collisions (my roads are really winding with low visibilty when curving left). Now I can do more testing on these roads. Car still cuts corners on left curves but it seems to realize this sooner and moves back over.

- car detects what type of lane faster. This is in-line with the release notes where the car is better at line marking interpretation. Perhaps this also explains that the car does a bit better with rotaries/roundabouts. Still needs refinement, but improving.

- I haven't seen much improvement with phantom breaking.

- creep is still way too slow for standard driving behavior around here. I always use the accelerator if someone is behind me. Car slows/stops way too early for stop signs. I've yet to experience the car stopping late or overshooting the stop, as others have reported.

- some seemingly easy turns take forever, or the car gets stuck and doesn't move.

 

[EVNow]

- less jerk/jitter with the steering wheel for turns. It's actually kinda black/white: either it turns very smoothly, or it's really erratic. I haven't encountered something in between.

One thing I’ve noticed is - if the car waits for another car to pass before turning - it is smoother. I wonder whether all the jittery behavior, esp when turning from a stop is because of the time it takes FSD to figure what’s happening at an intersection, figure out the distance and speed of any vehicles nearby. If it is waiting for another car to pass, it has had time to do all the figuring out.

 

[JulienW]

Voted Slightly Better which is modus operandi for a x.1 software update. In general it is a little smoother and lots of small improvements. Only regressions I find is it seems to be overall less assertive (which can also be good when it is doing the WRONG thing). Overall still find FB about the same. And don't see turning into traffic safely being improved at all. Still will go for it if it can't see or even if traffic is coming and is super cautious when the coast is wide open clear.

 

[powertoold]

Braking gradient while approaching / around lights is worse for me; otherwise, much fewer disengagements. With 10.4, I would disengage somewhat often because of curbing fear. It's much improved with 10.5, although 10.5 is less comfortable in a lot of ways.

At the end of the day, fewer disengagements is what really matters. Many YouTubers are reporting more zero disengagement drives, which is a significant improvement.

Hard to put a number on it, but perhaps 10.5 has 60% fewer disengagements vs 10.4?

 

[mark95476]

This has only happened once, but 10.5 made a rather dramatic maneuver to center itself in a wide (merge) lane. I had to disengage, but it was a very good drive otherwise. That's never happened before on the same road.

When I'm in the left/passing lane, 10.5 still wants/tries to go around slower traffic by switching to the right lane. I turned "exit passing lane" off in the settings and I need to approve all lane changes.

 

[mark95476]

10.4 and 10.5 both miss the first speed bump on Almond right by Los Alto High. It detects and slows for the second at Almond Elementary.

 

[mark95476]

This was weird and never happened before. FSDbeta 10.5 is driving, but map disappeared, and car started panicking and forced me to eventually take over while braking for red light and stopped traffic. Maps came back soon after and all was kosher. The LTE network is severely overloaded in the Bay Area and pushing 10.4 & 10.5 over LTE is probably straining things further.

BTW, braking with 10.5 is improved and feels natural. 10.5 seems to look further ahead and doesn't mindlessly speed-up to set speed only to brake hard for stopped traffic or lights.

 

[JulienW]

This was weird and never happened before. FSDbeta 10.5 is driving, but map disappeared, and car started panicking and forced me to eventually take over while braking for red light and stopped traffic. Maps came back soon after and all was kosher. The LTE network is severely overloaded in the Bay Area and pushing 10.4 & 10.5 over LTE is probably straining things further.

BTW, braking with 10.5 is improved and feels natural. 10.5 seems to look further ahead and doesn't mindlessly speed-up to set speed only to brake hard for stopped traffic or lights.

Not a big deal. That is just Google maps that is dynamically downloaded and used as a driver display. The car uses the Tesla Navigation database that is downloaded to the car. You don't see this. It is only a problem if it is not reading the Navigation database correctly.

Here is what the car uses to navigate.

 

[Phlier]

You would think that this would be an easy poll to respond to, but for me at least, it wasn't.

I've had 10.5 do more dangerous things than 10.4 did, by far. It's also done some really stupid (but not necessarily dangerous) things that 10.4 didn't do.

Yet if I look at it overall, 10.5 is a small improvement over 10.4.

Just yesterday, it tried to turn left at the street before it was supposed to turn left (it should have turned left at a traffic signal, but instead turned left at a residential street just before the traffic light). It started to turn left, but a car turned right at the intersection ahead (making it head on to me), and FSD just gave up... right in the middle of the opposing direction traffic lane. It took me nearly zero time to correct FSD's error, but the guy still looked at me like he'd just smelled a fart.

IMO, there's an occasional GPS accuracy bug that rears its ugly head from time to time, and with varying degrees of error... sometimes it shows me miles away from where I really am, and other times it's a very short distance (as was the case here).

I really think that a lot of 10.5's problems stem from some bad GPS code throwing off the whole thing.

 

[EVNow]

Not a big deal. That is just Google maps that is dynamically downloaded and used as a driver display. The car uses the Tesla Navigation database that is downloaded to the car. You don't see this. It is only a problem if it is not reading the Navigation database correctly.

Here is what the car uses to navigate.

View attachment 739833

I wish they just showed the map that the car uses. Atleast we can figure out easily if its a mapping error and ask them to correct. Now we have to go sleuthing to figure out exactly what is happening. WYSIWYG.

 

[JulienW]

I wish they just showed the map that the car uses. Atleast we can figure out easily if its a mapping error and ask them to correct. Now we have to go sleuthing to figure out exactly what is happening. WYSIWYG.

Amen to that. Also I bet 99.9% of Tesla drivers think that the way they see it on the display is exactly what the car sees and is responding to.

[wild speculation] I bet Tesla is buying and using data from different sources (TomTom probably a large part). Since this data is all "mashed" together for the Tesla Navigation database they probably aren't licensed to use it all visually together. So it is just easy for Tesla to overlay Google Maps for the user display.

 

[EVNow]

[wild speculation] I bet Tesla is buying and using data from different sources (TomTom probably a large part). Since this data is all "mashed" together for the Tesla Navigation database they probably aren't licensed to use it all visually together. So it is just easy for Tesla to overlay Google Maps for the user display.

This might be one of those Bay area issues. They probably notice and correct all the mapping issues in Bay area very frequently and are not aware how many issues crop up in other areas because of incorrect mapping.

I've also thought that Tesla does integration of maps from different sources. But all the issues I've noticed (and those people have written here about) seem to point to TomTom maps. So, not so sure about integrating with different sources any more - may be they just add the charger information from internal sources.

 

[Mardak]

10.4 and 10.5 both miss the first speed bump on Almond right by Los Alto High. It detects and slows for the second at Almond Elementary.

View attachment 738661

Is this a new speed bump directly south of the track/field as you've marked on the map? Seems like that's not in OSM, but there is a "speed table" (wider hump with crosswalk) mapped at the elementary school:

There's also some mapped speed humps on Jardin and Clark just north of the elementary school if you want to test.

 

[mark95476]

Yes, it's a speed table like the one by the elementary school and it's located right by the student drop-off entrance.

Thanks for checking into this. I need to create an account on OSM, TomTom, etc. and try fixing this stuff to see if it makes a difference.

BTW, FSDbeta 10.5 through Los Altos was very good for me and put a smile on my face. I didn't use NoA and FSDbeta was making the right turns at T intersections/stops/lights, etc. to get me where I wanted. 10.3/10.4 used to struggle at intersections and freak-out (audible warning with wandering).

Is this a new speed bump directly south of the track/field as you've marked on the map? Seems like that's not in OSM, but there is a "speed table" (wider hump with crosswalk) mapped at the elementary school:
View attachment 739864

There's also some mapped speed humps on Jardin and Clark just north of the elementary school if you want to test.