FSD v12.x (end to end AI)

[gsmith123]

How do you guys get these camera views? Is this available for all cars, or you guys have software that has hacked it ?

Settings > Service > Camera Preview. Available on all modern vehicles, I believe. The annoying thing is you have to be in park though so if you're curious what the various cameras can see while driving no luck

 

[KLam]

I think that is just my iPhone trying to capture it and not the is dash orientstion

Another way to check is to find a well marked parking lot. Park your car straight in the middle of the parking lines and check your rear views.

 

[FSDtester#1]

Settings > Service > Camera Preview. Available on all modern vehicles, I believe. The annoying thing is you have to be in park though so if you're curious what the various cameras can see while driving no luck

Correct, and you can also see all the cameras in the app when Sentry mode is on.
You can see multiple at once, or go to any individual camera.

 

[rlsd]

This thread was just a bit tired with everyone waiting for v12.4 videos to drop. Aren't you glad I brought up the B-pillar camera issue so we'd have something different to discuss . In a few days (hopefully) it will be v12.4 all the time!

It will be more fun when we discuss the rear camera.

 

[gottagofast]

The cameras getting dirty on the front bumper makes sense esp with the bugs I get

 

[aronth5]

While we wait for v12.4 my major list of improvements that get FSD "closer" to unsupervised.
These represent almost all my disengagements and are the most likely IMO to prevent FSD from reaching it's destination. The first five could easily result in moving violations so Tesla should focus on these IMO. Most other disengagements if skipped by the driver wouldn't prevent FSD from reaching it's destination. Might not be the smoothest drive but FSD would get there without an accident. I'm of the school to avoid disengagements if I can.

• Hand gestures by police/road crews
• School zones (in my town this is working reasonably well using the flashing light signs but there is so much variability from town to town)
• School buses with flashing lights
• Emergency vehicles. In my state it's the law to pull over.
• Improved speed control especially reducing speeds quickly when the legal speed limit goes down

Lower priority would be

• Pothole avoidance. Most are filled in now so this was a priority earlier this year
• Backing up
• Lane selection/fix wobbling

FSD is pretty close to disengagement free drives at least for me. I did have a critical safety engagement this past week. First in 1,845 miles.
Still no curb rash with v12.
Use auto max on 100% of drives. My advantage I think is there are no roads in my area that have a speed limit of 50 or 55mph. They are either controlled access highway or have speed limits between 25mph and 45mph.

 

[sleepydoc]

It will be more fun when we discuss the rear camera.

Ask @FSDtester#1 for some pics on Tuesday…

 

[sleepydoc]

The cameras getting dirty on the front bumper makes sense esp with the bugs I get

Elon said they’d be fixed with 12.4

 

[JHCCAZ]

It will be more fun when we discuss the rear camera.

Ask @FSDtester#1 for some pics on Tuesday…

Not if he gets Full Self Colonoscopy, then he won't need you anymore.
(He told me he's hoping it'll be a very big Release.)

 

[David Wang]

While we wait for v12.4 my major list of improvements that get FSD "closer" to unsupervised.
These represent almost all my disengagements and are the most likely IMO to prevent FSD from reaching it's destination. The first five could easily result in moving violations so Tesla should focus on these IMO. Most other disengagements if skipped by the driver wouldn't prevent FSD from reaching it's destination. Might not be the smoothest drive but FSD would get there without an accident. I'm of the school to avoid disengagements if I can.

• Hand gestures by police/road crews
• School zones (in my town this is working reasonably well using the flashing light signs but there is so much variability from town to town)
• School buses with flashing lights
• Emergency vehicles. In my state it's the law to pull over.
• Improved speed control especially reducing speeds quickly when the legal speed limit goes down

Lower priority would be

• Pothole avoidance. Most are filled in now so this was a priority earlier this year
• Backing up
• Lane selection/fix wobbling

FSD is pretty close to disengagement free drives at least for me. I did have a critical safety engagement this past week. First in 1,845 miles.
Still no curb rash with v12.
Use auto max on 100% of drives. My advantage I think is there are no roads in my area that have a speed limit of 50 or 55mph. They are either controlled access highway or have speed limits between 25mph and 45mph.

The biggest issue for me is still incorrect map data. I still run into situations where it tries to go straight in a left turning lane and vice versa.

This is my number 1 source of disengagements.

There must be a way for the AI to read the lane markings and override the incorrect map data.

Other than that I'm pretty happy w 12.3.6 and I'm using it w every drive.

The fsd shimmy is annoying but I've gotten used to it now so it's not a big deal anymore.

The hesitation at stops is also annoying but Ive learned to push on the accelerator if there are cars behind me, if not I just let it do it's thing.

 

[RDoc]

While we wait for v12.4 my major list of improvements that get FSD "closer" to unsupervised.
These represent almost all my disengagements and are the most likely IMO to prevent FSD from reaching it's destination. The first five could easily result in moving violations so Tesla should focus on these IMO. Most other disengagements if skipped by the driver wouldn't prevent FSD from reaching it's destination. Might not be the smoothest drive but FSD would get there without an accident. I'm of the school to avoid disengagements if I can.

• Hand gestures by police/road crews
• School zones (in my town this is working reasonably well using the flashing light signs but there is so much variability from town to town)
• School buses with flashing lights
• Emergency vehicles. In my state it's the law to pull over.
• Improved speed control especially reducing speeds quickly when the legal speed limit goes down

Lower priority would be

• Pothole avoidance. Most are filled in now so this was a priority earlier this year
• Backing up
• Lane selection/fix wobbling

FSD is pretty close to disengagement free drives at least for me. I did have a critical safety engagement this past week. First in 1,845 miles.
Still no curb rash with v12.
Use auto max on 100% of drives. My advantage I think is there are no roads in my area that have a speed limit of 50 or 55mph. They are either controlled access highway or have speed limits between 25mph and 45mph.

No Right Turn on Red signs!

 

[hey8you]

The biggest issue for me is still incorrect map data. I still run into situations where it tries to go straight in a left turning lane and vice versa.

This is my number 1 source of disengagements.

There must be a way for the AI to read the lane markings and override the incorrect map data.

Other than that I'm pretty happy w 12.3.6 and I'm using it w every drive.

The fsd shimmy is annoying but I've gotten used to it now so it's not a big deal anymore.

The hesitation at stops is also annoying but Ive learned to push on the accelerator if there are cars behind me, if not I just let it do it's thing.

Same here. I don't know how often and what data gets updated but the "no turn on red" at one intersection near me is 6 years old and where v12 inappropriately enters the wrong lane during a turn on another intersection is 13+ years old.

 

[jabloomf1230]

The cameras getting dirty on the front bumper makes sense esp with the bugs I get

These cameras would benefit left and right turns. The "bumper" cameras would be incorporated into each headlight assembly and would face at a 90° angle to the direction of the vehicle. It's not a hard problem to solve.

 

[Ben W]

The cameras getting dirty on the front bumper makes sense esp with the bugs I get

Elon said they’d be fixed with 12.4

They'd be squashed, at any rate.

 

[Ben W]

These cameras would benefit left and right turns. The "bumper" cameras would be incorporated into each headlight assembly and would face at a 90° angle to the direction of the vehicle. It's not a hard problem to solve.

Considering how much glare/interference there already is between repeater cameras and turn signal blinkers, I wonder how they would avoid the glare problem at night if the bumper cameras are literally inside the much brighter headlight assembly. Or in rainy conditions, brightly lit (by the headlights) nearby raindrops might completely obscure far-away details. So it may end up being necessary to offset the bumper cameras by at least a foot or so from the headlights per se, or perhaps to use a more exotic solution, like IR cameras that detect a wavelength less "polluted" by the headlights.

 

[JHCCAZ]

These cameras would benefit left and right turns. The "bumper" cameras would be incorporated into each headlight assembly and would face at a 90° angle to the direction of the vehicle. It's not a hard problem to solve.

I've long advocated cameras in the headlight assemblies, but I don't think it's only for 90° side view. Just as important is the ability to get the best angle to look at the high speed oncoming lanes, past left-turning vehicles, signs in the median or other obstructions. The forward view of oncoming traffic, from a camera placed at or near the left fender, is far superior to the forward view from the existing forward facing cameras mounted in the center of the windshield.

Considering the headlight or similar front corner location, a camera with a 110° view, angled left at 45°, would take in the side view and the front view.

Another way to do this, which Tesla seemed to be moving towards with early HW4 prototypes, is to include a forward-looking camera in the repeater as a companion to the existing rearward-looking camera. For side viewing it's not as far forward as the headlight, but it's still well forward of the driver's view. There are other possible variations, for example front and rear looking cameras at the current wing mirror location (even though the mirror itself becomes obsolete). It looks like Tesla may be experimenting with that also.

 

[JHCCAZ]

Considering how much glare/interference there already is between repeater cameras and turn signal blinkers, I wonder how they would avoid the glare problem at night if the bumper cameras are literally inside the much brighter headlight assembly. Or in rainy conditions, brightly lit (by the headlights) nearby raindrops might completely obscure far-away details. So it may end up being necessary to offset the bumper cameras by at least a foot or so from the headlights per se, or perhaps to use a more exotic solution, like IR cameras that detect a wavelength less "polluted" by the headlights.

Early last year when HW4 prototypes and rumors were around, I speculatively addressed this issue of how the housing assembly should be designed to allow glare-free optical input for the camera along with high power illumination output for the headlight:

I agree that this is very unlikely to be an hw4 car.

I've long suggested the idea of cameras in the headlight and/or taillight assemblies. The motivation for this argument is a) it's a pretty good spot for corner cameras, b) it would have shared the body opening and potentially wire harness routing for retrofit purposes, and c) there's an easy path to incorporate a washer system to service cameras as well as headlights.

However, in such a design the cameras would not be placed back where the LED emitters are, put in a separate sub-housing right up against the outer lens. Otherwise yes, there would be a huge glare problem.

In my onception, it would be more akin to the camera assembly directly behind the windshield but within its own housing. Going a bit further though, I believe it should not even share the outer shell polycarbonate lens of the headlight, but would have its own separate outer lens with a minimal opaque border for separation. This would further prevent oblique surface glare transfer from a dirty or scratched outer headlight lens.

I mention all the details just above because it's an interesting design problem and IMO solvable at reasonable cost. But to be clear, and as happy as I'd be to see it, I'm very doubtful that Tesla has actually chosen this approach of cameras within the headlights.

We'll see soon, but I don't think this is it.

 

[FSDtester#1]

I just wanted to say congratulations to @LSDTester#1
, formerly known as pdubs, as being the first known Global Lsd expert in the field of all things strange, weird, and exciting! Great job @LSDTester#1

If you keep this rate of progress, maybe, just maybe, you can be Robin one day!

 

[LSDTester#1]

I just wanted to say congratulations to @LSDTester#1
, formerly known as pdubs, as being the first known Global Lsd expert in the field of all things strange, weird, and exciting! Great job @LSDTester#1

If you keep this rate of progress, maybe, just maybe, you can be Robin one day!

Thank you one and all. None of this would be possible without my dear friend and mentor Timothy Leary

 

[sleepydoc]

While we wait for v12.4 my major list of improvements that get FSD "closer" to unsupervised.
These represent almost all my disengagements and are the most likely IMO to prevent FSD from reaching it's destination. The first five could easily result in moving violations so Tesla should focus on these IMO. Most other disengagements if skipped by the driver wouldn't prevent FSD from reaching it's destination. Might not be the smoothest drive but FSD would get there without an accident. I'm of the school to avoid disengagements if I can.

• Hand gestures by police/road crews
• School zones (in my town this is working reasonably well using the flashing light signs but there is so much variability from town to town)
• School buses with flashing lights
• Emergency vehicles. In my state it's the law to pull over.
• Improved speed control especially reducing speeds quickly when the legal speed limit goes down

Lower priority would be

• Pothole avoidance. Most are filled in now so this was a priority earlier this year
• Backing up
• Lane selection/fix wobbling

FSD is pretty close to disengagement free drives at least for me. I did have a critical safety engagement this past week. First in 1,845 miles.
Still no curb rash with v12.
Use auto max on 100% of drives. My advantage I think is there are no roads in my area that have a speed limit of 50 or 55mph. They are either controlled access highway or have speed limits between 25mph and 45mph.

• Hand gestures by road crews - Virtually never have this. Not sure why it should be a priority
• school zones - discussed here before. There is no consistency, even within a single town or school district, and humans are frequently unsure how fast they can drive.
• School busses - absolutely. Shouldn't be too tough.
• Emergency vehicles - FSD used to 'recognize' emergency lights but it was virtually always wrong. Haven't seen it recently, though.
• Peed control - this should be the first 5 items on the list.