2022.20.9 - Tesla Vision impressions

[Mark II]

With respect to Radar vs Vision. I have a 2018 with Radar and a 2022 without radar, i.e. vision only. Two main things I have noticed:
1 - For a chosen setting the following distance of Radar is much more consistent, versus Vision that seems more squishy. Although not a huge deal.
2 - The biggest difference I have seen with AP without radar is that it only reacts to the car directly in front of me. I.e. the following distance on freeway with no radar does not account for another car in front of the car that is directly in front of me. With the 2018, if the 1st car directly in front of me was too close to the 2nd car in front of it the following distance would increase. Also if the 2nd car started to brake my 2018 would start braking even before the 1st car directly in front of me started to brake. Which overall felt safer to me.

 

[SoCal Buzz]

Just completed same 80 mile trip I do most weeks on Tesla Vision for first time (2021+ Refresh on 2022.20.9). AP was smooth with no phantom braking or other issues, both in stop and go freeway traffic and at 75 mph+ for about 1/2 of the trip. I noticed very little difference in behavior, except perhaps more graceful slowing in traffic. It still takes time to get back up to speed, so no change there. Overall I would say slight net + but not much difference so far.

My biggest issue with AP in general is when single lanes split into 2 (e.g., carpool passing lane) or when lane markers fade out, the car tends to jerk one way or the other trying to find the lane. Removing radar from the equation won't affect that behavior, so hoping vision improvements lead to smoother transitions and handling with poor lane markers.

 

[mtndrew1]

Just tonight, on a <10 mile drive on I-25 through town, the 2022 MSLR had a relatively mild phantom braking (probably only lost about 5MPH before I got on the accelerator), fortunately with nobody near. I filed a bug report, but I suspect those are being ignored until the magic day in the far-off future when we get to test "single stack" AP.

So, no, it doesn't appear that 2022.24.5 offers any improvement. Not the I expected it would.

Bug reports are only used in conjunction with active service center visits. Nobody at Tesla pays them any attention otherwise so unless you have a service ticket open, don’t bother.

 

[ShaneT]

Just completed same 80 mile trip I do most weeks on Tesla Vision for first time (2021+ Refresh on 2022.20.9). AP was smooth with no phantom braking or other issues, both in stop and go freeway traffic and at 75 mph+ for about 1/2 of the trip. I noticed very little difference in behavior, except perhaps more graceful slowing in traffic. It still takes time to get back up to speed, so no change there. Overall I would say slight net + but not much difference so far.

My biggest issue with AP in general is when single lanes split into 2 (e.g., carpool passing lane) or when lane markers fade out, the car tends to jerk one way or the other trying to find the lane. Removing radar from the equation won't affect that behavior, so hoping vision improvements lead to smoother transitions and handling with poor lane markers.

Same experience for me. I had 2022.20.8 for my 2017 P100D with the upgraded MCU. I have been noticing that the braking has been a bit smoother even with .8 but no major difference with .9. It seems to operate very smoothly with no phantom braking. One thing I have noticed is less back and forth on the steering wheel during long curves in the road (Nothing official but just my opinion). I would give 2022.20.9 a thumbs up so far!

 

[hemanthpai]

I am holding out as long as I can. I had one of the original Model Y's when they dropped the radar. Teslavision was total crap. I patiently waited almost a year for it to get better. I sold the car and the current owner still has phantom braking (PB) issues.

I'll add another data point. I have a 2022 Model 3 (without radar). It is better than the Y was but nowhere near as good as the S has been. I live in Texas and dropping the max speed when on AP to 85 mph totally sucks. The problem is the speedo isn't accurate so 85 isn't even a true 85 mph. If you go on Texas State Highway 130, the speed limit is 85. What a cluster this is.

My Y had extreme PB issues. I drive on the same roads with the S and in 6k miles I've only only ONE PB issue in its entire life which must be some kind of miracle. My Y was had probably close to 100 on one trip from FL to TX. My S traveled the same roads and not a single issue. I credit the radar for most of that. The 3 isn't psychotic like the Y was but it doesn't it just often enough to make me want to cover the accelerator in case it brakes. The Y would almost slam on the brakes while driving on just TACC at 85 and drop you to 50 in a heartbeat it seemed.

So I am holding off on this update likely as long as I can or until I see irrefutable evidence that it won't cripple my car like the Y. I'll still take the adaptive cruise control in a 9 year old Toyota Avalon over the Teslavision option for just cruise. It is like because of Tesla the term "phantom braking" was coined so to speak. I drive probably close to 50 different rental cars each year and nobody's adaptive cruise was as bad as I experienced in the Y. The 3 is clearly not in the top tier. I know the Tesla Stans will still be drinking the Kool Aid about this but when Tesla can't even figure out not to drive the car forward with the auto direction select when there is a wall in front of the car, then I it is obvious they have issues with some of the basic. Sure, it is likely a different team working on but if you can mess up on such a major issues such as your parking sensors say there is an object in front of the car, then maybe you should select "forward" for the direction of travel.

Good luck to all of you rolling the dice on this and going with it. I'll wait a few releases and see if they get the bugs worked out.

I have a 2020 Model X (17K miles) that has radar and I experience phantom braking a couple times a month (10s of times over the entire mileage I’ve driven). Totally agree that Tesla’s Adaptive Cruise Control is worse than other, older cars that offer Adaptive Cruise Control. I will also be holding off on updating for as long as I can. Can’t believe Tesla is « forcing » this update on us vs letting us choose if we want to make the switch.

 

[Waldokind]

Thanks for the detailed report! How was the night driving? Did you have to disable auto high beams every time you enable the AP/FSD?

I have not driven at night yet. My windshield wipers were already set on auto (works fine for me in SoCal) and that's unchanged. My high beams are still in manual mode but maybe that will change when I'm actually in the dark? Dunno yet.

 

[joexn]

Tesla Vision is coming to cars with 2022.24.6

https://twitter.com/x/status/1565053701038940162

 

[MikeyC]

Same experience for me. I had 2022.20.8 for my 2017 P100D with the upgraded MCU. I have been noticing that the braking has been a bit smoother even with .8 but no major difference with .9. It seems to operate very smoothly with no phantom braking. One thing I have noticed is less back and forth on the steering wheel during long curves in the road (Nothing official but just my opinion). I would give 2022.20.9 a thumbs up so far!

The above describes what I see in my 2021 Model 3 (USED to have radar) as well. I noticed a lot fewer micro-corrections when holding the center of the lane with 20.9. Thought I was imagining things but maybe not. I also noticed the car takes less time to recenter when there's an opening/gap in the lane lines and they resume at a slightly different width. Used to take a couple seconds to correct for the new lane width. Faster now. Wouldn't have thought vision (no radar) would affect that but if it comes with a new branch of AP, it's possible the vision branch of AP behaves slightly differently.

Mike

 

[ShaneT]

The above describes what I see in my 2021 Model 3 (USED to have radar) as well. I noticed a lot fewer micro-corrections when holding the center of the lane with 20.9. Thought I was imagining things but maybe not. I also noticed the car takes less time to recenter when there's an opening/gap in the lane lines and they resume at a slightly different width. Used to take a couple seconds to correct for the new lane width. Faster now. Wouldn't have thought vision (no radar) would affect that but if it comes with a new branch of AP, it's possible the vision branch of AP behaves slightly differently.

Mike

I’m with you… doubt it has anything with removing radar. I think we got some upgraded AP code with the release.

 

[Waldokind]

I have not driven at night yet. My windshield wipers were already set on auto (works fine for me in SoCal) and that's unchanged. My high beams are still in manual mode but maybe that will change when I'm actually in the dark? Dunno yet.

I can confirm engaging autopilot with this new patch does not activate automatic high beams. My high beams were on manual setting, and off, and after driving an hour in the dark turning AP on and off, my high beams are still in manual.

 

[Rice&Curry]

I would not use the FSD in a snow storm (it always wants to go with the speed limit) but I would use the road picture when the road is covered with snow and I cannot see the lanes. Can the car see the lanes with Tesla Vision only?

I am not sure if the radar can detect the lanes if the road is covered in snow , so in this particular situation radar or no radar the lane markings most probably won't be detected.

 

[joexn]

I am not sure if the radar can detect the lanes if the road is covered in snow , so in this particular situation radar or no radar the lane markings most probably won't be detected.

Yeah radar can see objects, not painted lines.

 

[Boza]

Yeah radar can see objects, not painted lines.

I know and that was why I am wondering how the car was able to see lanes when I could not. Wondering if the paint is more radar reflective than the asphalt on the road. If we could go the other way (creating stealth paint that absorbs the radar) there should be a way to have more reflective paint, albeit not intentionally.

 

[Rice&Curry]

Both my cars MX and MY skipped 2022.20.9 and went to 2022.24.6

 

[MikeyC]

I've had maybe 100 miles on 2022.20.9 in my Model 3 and I can say that I see no difference whatsoever with radar versions before it. The one thing I'm still on the fence about might be cross traffic. You know the scenario: you're driving down a straight road and someone way down the road makes a left turn across your lane and the car freaks out and slams on the brakes even though YOU can tell they will be out of your lane long before you get there. This behavior MIGHT be slightly improved. Hard to tell because it does still do it.

The thing they really need to fix is rural roads where a bike lane opens up on the right. Those add a second white line on the right maybe 3 feet farther to the right and when that happens, AP swaps between the lane line, then bike lane line, then lane line, then back to bike lane line (for the right edge of the lane) and this "flicking" of the width of the lane makes the car swerve back and forth quickly and many times, causes it to slam on the brakes. This is in dire need of a fix as it's been there as long as I've owned my Model 3 (a year and a half). All they have to do to fix it is just consider the closest white line to the right to be the right edge of the lane: don't get confused and pick the second line sometimes even when all lines are highly visible/new lines.

Mike

 

[Boza]

I've had maybe 100 miles on 2022.20.9 in my Model 3 and I can say that I see no difference whatsoever with radar versions before it. The one thing I'm still on the fence about might be cross traffic. You know the scenario: you're driving down a straight road and someone way down the road makes a left turn across your lane and the car freaks out and slams on the brakes even though YOU can tell they will be out of your lane long before you get there. This behavior MIGHT be slightly improved. Hard to tell because it does still do it.

The thing they really need to fix is rural roads where a bike lane opens up on the right. Those add a second white line on the right maybe 3 feet farther to the right and when that happens, AP swaps between the lane line, then bike lane line, then lane line, then back to bike lane line (for the right edge of the lane) and this "flicking" of the width of the lane makes the car swerve back and forth quickly and many times, causes it to slam on the brakes. This is in dire need of a fix as it's been there as long as I've owned my Model 3 (a year and a half). All they have to do to fix it is just consider the closest white line to the right to be the right edge of the lane: don't get confused and pick the second line sometimes even when all lines are highly visible/new lines.

Mike

Can you pls elaborate on the conditions - weather, traffic, time of day, etc? My concern is that cameras will be OK most of the time but in critical situations they will perform worse than camera + radar.

 

[MikeyC]

Can you pls elaborate on the conditions - weather, traffic, time of day, etc? My concern is that cameras will be OK most of the time but in critical situations they will perform worse than camera + radar.

In my case, time of day or conditions don't matter. The issue I was outlining is misinterpretting the lines on the road so radar is not involved there. It is possible that vision+radar could detect objects on the road at a greater distance in poor visibility but that's another topic altogether. In such cases, I would argue that radar is "superhuman", meaning we don't have radar when we drive. But that doesn't necessarily mean it is "wrong" to use a technology that could make for better-than-human detection so I'm not against radar. If used, radar should probably be limited to those cases where poor visibility exists to avoid potential false triggering from bridges and other obstacles that are not actually in the roadway.

Mike

 

[Boza]

In my case, time of day or conditions don't matter. The issue I was outlining is misinterpretting the lines on the road so radar is not involved there. It is possible that vision+radar could detect objects on the road at a greater distance in poor visibility but that's another topic altogether. In such cases, I would argue that radar is "superhuman", meaning we don't have radar when we drive. But that doesn't necessarily mean it is "wrong" to use a technology that could make for better-than-human detection so I'm not against radar. If used, radar should probably be limited to those cases where poor visibility exists to avoid potential false triggering from bridges and other obstacles that are not actually in the roadway.

Mike

Absolutely! Taking a page from other industries with long tail and critical functions. Usually, you have a number of (ideally) different sensors and computers compare and “vote” on what reality is.
In fact, what Tesla is doing reminds me a bit of 737Max - ignoring a sensor that is already available…

 

[sandman1330]

I’ve been holding off on FSD Beta because of worry about losing radar. Now that my car has queued up 2022.24.6, instead of accepting that update, since I’m going to lose radar anyway this is happening. Let the granny driving begin!

 

[MikeyC]

This is precisely why I'm not interested in the FSD beta. It is not worth me having to alter my driving to drive like a monk and be criticized by Tesla in order to get into the beta. I'd rather enjoy driving my car. When they release it to the masses and don't require "sterile" driving, I'll reconsider.

Mike