FSD v12.x (end to end AI)

[tmoz]

No I think he had it right.
Affect as a noun (pronounced AF-fect). A purposeful and somewhat over-expressed response or demeanor, put on for effect.

Like an over-the-top actor delivering a line or a comedian playing for the laugh.

Or on the internet, an emoji to communicate the tone - the affect.

It can be a good or a bad thing, but sometimes annoying. Sometimes when a person is prone to a put-on accent or demeanor, people will say that's an affectation.

I thought @APotatoGod used it purposefully and cleverly there. Good but uncommon word, and I know that the clever joke is diminished by having to explain it.

OK - it is good to learn something new everyday. @APotatoGod - intentional or accidental?

 

[EVNow]

I thought @APotatoGod used it purposefully and cleverly there.

Yes, that use was effective.

 

[PianoAl]

Can we get this on video with multiple cars doing this at the same location? How reproducible is this?

Not reproducible. In the post, i reported the car riding the center line on a straightaway. Yesterday, same road, same spot, it stayed in the lane perfectly.

However, there are curves where the car almost always drives over the line.

 

[aronth5]

Why UPS Trucks Don't Turn Left - Priceonomics

UPS engineers map out routes for their drivers consisting of "a series of right-hand loops"

priceonomics.com

I've seen similar articles as well. If Tesla decides to turn right instead of taking Chuck's complex UPL turn I'd be fine with that. That type of intersection is certainly one of the most difficult so just go up the street .2 miles and use the traffic light and turn left. (this option was presented by a driver who lives in the area)

 

[EVNow]

I've seen similar articles as well. If Tesla decides to turn right instead of taking Chuck's complex UPL turn I'd be fine with that. That type of intersection is certainly one of the most difficult so just go up the street .2 miles and use the traffic light and turn left. (this option was presented by a driver who lives in the area)

Yes, we have been through this a number of times. I almost always turn right if I can't quickly turn left (generally one traffic signal of the nearest signal). Otherwise someone at the back is going to start honking.

 

[enemji]

We build a world of laws because it better than than a world of brutes and emotional manifestations of frustration, possession or egotism…

If we define the real world requirements of replacing vehicle operation for transportation very carefully we may find ourselves with unexpected solutions

A large part of the unsatisfied people seem to gravitate towards this style of driving.

 

[enemji]

I've seen similar articles as well. If Tesla decides to turn right instead of taking Chuck's complex UPL turn I'd be fine with that. That type of intersection is certainly one of the most difficult so just go up the street .2 miles and use the traffic light and turn left. (this option was presented by a driver who lives in the area)

A right turn followed by a u-turn is best for safety.

In NJ this variant is called a jug handle type B turn.

 

[CyberGus]

A right turn followed by a u-turn is best for safety.

In NJ this variant is called a jug handle type B turn.

U-turns are illegal in Ohio, so I guess you would turn right and drive to Indiana?

 

[Daniel in SD]

Which evidence? The VTTI study seems to show that accident rates decrease roughly in proportion for all severity levels (for autonomous car vs human), at least for levels 1 - 3 in one of their charts. This seems to imply that e.g. halving the minor collision rate (through software/hardware advancement) would be likely to roughly halve the severe collision rates as well. Not in a direct causal sense of course, but because improving the overall software/hardware would tend to reduce mistakes at all levels of severity, and because such improvements tend to not be specifically targeted at narrow accident types (e.g. reducing curb strikes). (Some may be, but it's the exception.)

Waymo data from Phoenix showed a very different distribution from human drivers. In 6 million miles only 30 of least severe category (real and counterfactual simulation) and 8 with airbag deployment. Also note that there were zero collisions that did not involve another human.

https://assets.ctfassets.net/e6t5diu0txbw/7sFIWKmldD2A7tjYsJo5Ws/45a6b4169a5ccfece58683fe1d57777d/Waymo_Public_Road_Safety_Performance_Data.pdf

 

[EVNow]

I don't think there's anything profound, technically impressive, or otherwise appealing about substituting a left turn with 3 right turns.

If it reduces accidents it is worth it at a fleet level.

You can't dictate the route a robotaxi takes just like (most of the time) you can't dictate the route uber takes, or the UPS driver who drops off your package took

 

[EVNow]

Waymo data from Phoenix showed a very different distribution from human drivers. In 6 million miles only 30 of least severe category (real and counterfactual simulation) and 8 with airbag deployment. Also note that there were zero collisions that did not involve another human.
View attachment 1046746

https://assets.ctfassets.net/e6t5diu0txbw/7sFIWKmldD2A7tjYsJo5Ws/45a6b4169a5ccfece58683fe1d57777d/Waymo_Public_Road_Safety_Performance_Data.pdf

Even given the small numbers, you see decreasing numbers with severity.

BTW, I disagree small collisions don't matter. For ADAS it definitely does and even for robotaxi it shapes up the confidence and has PR implications (remember the when a robotaxi entered a freshly asphalted under construction road ?)

 

[aronth5]

U-turns are illegal in Ohio, so I guess you would turn right and drive to Indiana?

Just googled U-turns in Ohio and this statement is not accurate. There are lots of conditions in Ohio where U-turns are legal.
Besides the Jug handle B turn you questioned @enemji i is very different. Although not common we have a few in Mass as well. You have the option to go straight across or go left which in effect acts as a U-turn.

.

 

[EVNow]

U-turns are illegal in Ohio, so I guess you would turn right and drive to Indiana?

Hmm ...

Making a U-turn is not illegal in Ohio. However, there are some rules that drivers must follow before making a U-turn. Under the Ohio traffic code statute 4511.37, U-turns are prohibited on curves or hills where it may be impossible for other drivers to see the turning vehicle1. So, while U-turns are generally legal, it’s essential to consider the specific road conditions and visibility when making one. Keep those wheels turning safely!​

 

[sleepydoc]

It does a terrible job in busy parking lots.

In a parking garage, Navigation thought it was driving on the street outside the garage. I don't know what it would have done if I had enabled FSD, because it seemed like a bad idea to try it. This was exiting after an event, so traffic was bumper-to-bumper going down the ramps with no room for error.

On a 6-lane UPL, it waited so long to cross the near lanes, and moved so slowly, that it had to stop in the middle of them because before it reached the far lanes, traffic was approaching from the right.

Why would it need to go 15% above the limit at all times?

to be fair, parking lots are not roads, are poorly mapped, are often poorly marked and confusing anyway, and to my knowledge aren’t officially part of the designed domain for FSD. Add multiple layers of a parking ramp and it gets even worse.

If FSD gets me to the entrance I’m happy.

 

[CyberGus]

Just googled U-turns in Ohio and this statement is not accurate.

Making a U-turn is not illegal in Ohio. However...

In my defense, it's been many years since I was pulled over for "performing a U-turn in front of a police officer"

 

[sleepydoc]

LOL… You seem to have some bad eggs in your neighborhood grocery store!

FSD is simply following the rules, cautiously, as it should. Not sure why people are mad about it.

Worst time to go grocery shopping: 10 AM on weekdays. That’s when all the octogenarians are tottering down the middle of the aisles!

 

[sleepydoc]

Yep, sure enough recalibrating the cameras made a world of difference. It feels like a completely different system and is now very usable. It did take a VERY long time for the recalibration for the FSD, probably over 100kms of highway driving.

This is something Tesla will need to address eventually. Say they get the system with 1 accident every 10 million miles as long as the cameras are calibrated but the reliability drops to 1 accident every 1000 miles if they're not calibrated. People won't know and won't care if the issue was camera calibration. In the end, it's all part of the same system that they expect to work. Just like FSD needing to recognize when the weather is too poor for it to work reliably, it also needs some means of determining when the camera calibration is interfering with function.

 

[zoomer0056]

Yes, we have been through this a number of times. I almost always turn right if I can't quickly turn left (generally one traffic signal of the nearest signal). Otherwise someone at the back is going to start honking.

Why so? The person behind you should see you can't safely make the turn. Why would they honk and try to force you to drive beyond your ability or safety limits? On the other hand, ignore them and drive within your safe abilities. It's only in the movies that the impatient driver behind you will get out of their car with sledge hammer.

 

[aronth5]

With all the negativity by some about FSD this was refreshing

https://twitter.com/x/status/1789882064298193056

 

[lzolman]

Just like FSD needing to recognize when the weather is too poor for it to work reliably, it also needs some means of determining when the camera calibration is interfering with function.

Is there any known way to test calibration? I'd love to know if and how much my cameras are off before recalibrating them, and then to be able to measure again after.