2016 FSD video: there was an accident with roadside barrier on Tesla property

[apsen]

Aren't you even a bit alarmed by the fact that pure vision forces autobrights on?

And you never use brights when you driving yourself and can use them?

It was so bad that the car even slowed down under TACC only due to weather. It wasn't even raining that much.

I think you mean under autopilot... But how do you know it was due to Pure Vision? And if we are talking about single case scenarios, I just had 1500 miles trip with quite a bit of snow and rain and couldn't complain about its ability to drive in those conditions. My main complain is that they haven't brought back the following distance yet. I would have preferred to be able to increase it on the highways.

 

[apsen]

It comes down to how the accuracy of the vision system compares to Radar+Vision or Radar+Vision+Lidar.

Yes you could measure distances with better precision if you use lidar. But do you need it? What would you do with it?

The autonomy day video did an excellent job comparing Vision only with the old Vision+Radar system, but then they left out medium to long distances.

If someone leaves something on an otherwise excellent presentation it does make one wonder why?

How about the length of the presentation? Besides that was a recruiting event. And even if it was a scientific paper the size of it still matters. Also writing separate papers on medium and then on long distance would be a benefit as that would increase the number of publications.

The fact that its a work in progress does support that. Like right now its limited to 80mph versus 90mph of the radar+vision implementation. The fact that it hasn't reached parity means any comparison is a bit premature.

Yes, it seems they have rushed it a little bit. I suspect it was forced by the supplies issues.

At this point its not even about Lidar, but the question of whether we're going backwards.

I believe they did say that something like two step forwards one step back. It is the nature of neural networks training.

 

[S4WRXTTCS]

I think you mean under autopilot... But how do you know it was due to Pure Vision? And if we are talking about single case scenarios, I just had 1500 miles trip with quite a bit of snow and rain and couldn't complain about its ability to drive in those conditions. My main complain is that they haven't brought back the following distance yet. I would have preferred to be able to increase it on the highways.

No this was specifically under TACC, and I was using TACC because I got a similar message under AP.

I know its due to PureVision because its the first time I've gotten this message since having the car, and being in Seattle I have a lot of experience driving in the rain.

The car I own is a 2018 Model 3 with HW3/FSD, and the PureVision came with the FSD Beta. One of the known drawbacks of doing the FSD Beta upgrade is that you have to move to Pure Vision.

This isn't just my own experience, but its pretty widely accepted that PureVision doesn't perform in bad weather as well as the previous Vision+Radar. This makes sense as the Radar does fine in the rain.

 

[S4WRXTTCS]

And you never use brights when you driving yourself and can use them?

Most drivers use brights when the situation calls for them.

If I'm driving down a rural road late at night then I'll use brights to improve visibility. But, if I'm driving in a metro area right after the sun has gone down I'm not going to use brights.

The problem with PureVision is it doesn't use any logic in turning on the autobrights. If the lights are on it automatically turns on the autobright system whenever AP/FSD is in use.

When the situation doesn't warrant it I immediately turn them off.

In my area there are a lot of Tesla's so I regularly see one with their autobrights on, and its pretty comical. It's comical because they're constantly turning on and off.

 

[AndreP]

Autobrights are pretty flawed in a lot of vehicles tbh, but they're usable enough in my daily drivers (non-Tesla) that I let them do their thing.

But they activate and turn off at odd or inappropriate times, don't function well or account for stuff like heavy snow or fog, and other issues. This is how new such technology is, legitimately can't even flawlessly perform one of the simplest functions in an automobile. The light sensors are generally good at ensuring you don't forget they're on and end up blinding oncoming drivers, but they're still not flawless even there.

 

[daktari]

I hope you americans soon can experience modern matrix lights. That is really amazing tech.

 

[mark95476]

Out of context, but whatever...

We all know that MobilEye had a very public break-up with Tesla. Mobileye's CEO still seems butt-hurt and can't shut-up about Tesla years later.

I agree with this statement, but I'm confused by its source.

The only one pointing the finger at the vendor (MobileEye) is you.

You accused MobileEye for a weakness in the Ford Blue Cruise system. Whether are not that weakness is due to MobileEye is irrelevant to the customer. I only chimed in because I felt like you didn't understand how ADAS systems are implemented into vehicles. That there is often a lot of decisions up to manufactures in how features will work, and what's supported. There can also be a lot of customization.

 

[apsen]

No this was specifically under TACC, and I was using TACC because I got a similar message under AP.

Could you provide more details on that? I do not understand why would TACC slow you down without a car in front. Admittedly, I almost never use the TACC since if the autopilot/autosteer is not available and I have to steer it would be either short stretches or places where you can't really use cruise control anyway and in that case I could as well control the speed.

 

[apsen]

The problem with PureVision is it doesn't use any logic in turning on the autobrights. If the lights are on it automatically turns on the autobright system whenever AP/FSD is in use.

So because the autobrights logic is far from perfect it means AP/FSD is flawed?

 

[S4WRXTTCS]

Could you provide more details on that? I do not understand why would TACC slow you down without a car in front. Admittedly, I almost never use the TACC since if the autopilot/autosteer is not available and I have to steer it would be either short stretches or places where you can't really use cruise control anyway and in that case I could as well control the speed.

It specifically said it was slowing the set speed due to weather. That was the first time I had ever seen any Tesla Adaptive cruise control system slow due to moderate rain. If it was heavy rain I wouldn't have been going that fast or if there was a lot of standing rain.

This experience was with AP/TACC with the FSD Beta 10.5 firmware, and during the course of this 400+ mile journey I experienced all sorts of weather related weirdness that I had never experienced with my car before the FSD Beta.

I'm not ruling out that FSD Beta might be more sensitive to rain or cameras blinded. They've clearly been making adjustments with detecting weather.

 

[S4WRXTTCS]

So because the autobrights logic is far from perfect it means AP/FSD is flawed?

What's flawed is the reliance on autobrights.

I have absolutely no issue with Pure Vision wanting to control brights. If it had control over turning on the bright I would simply observe when it did, and to make sure it was doing so only when necessary. But, it doesn't actually do that. It simply turns on autobrights, and gives no indication to the user whether brights are even necessary. I've gotten into the habit of always forcing autobrights back off after enabling FSD/AP.

Is turning it off impacting how FSD Beta performs? I don't know.

It also tells me "FSD degraded due to weather or blinded camera" a LOT and who knows if it really is or not.

The bottom line is its very much still in development.

It's not really even about lidar.

What its about Elon promised FSD, and in doing made it cost prohibitive to make any sensor suite change.

The HW hasn't been allowed to grow, and as a result SW is forced to make due with the existing sensors.

Probably the coolest feature of FSD Beta is when actives the windshield cleaning. Too bad it can't do that with all the other cameras. Or when it says "camera blinded" that it can't put on sunglasses (as in inserting an ND filter).

 

[dduffey]

However, I tend to agree with several of the criticisms in there. Copilot would've been a better name than Autopilot.

I haven't read the article (paywall) but how is "Copilot" a better name than "Autopilot"?

I've heard this a few times that "Autopilot" should be called "Copilot" but to me that's backwards. By definition a co-pilot is a fully functional second pilot where you can pass off control and go to the bathroom, read a book, or take a nap. "Autopilot" is "a device for maintaining a set course" and does not allow a pilot to do the above (i.e. a pilot still needs to be engaged and aware). So it appears to me that "Autopilot" is the correct term to express these limits/expectations, not Copilot.

FSD should be called "Copilot" if anything (i.e. A Co-pilot is more capable than an auto-pilot, not the other way around). Also in FSD Beta and agree it's not full-self-driving nor Copilot capable, yet? .

 

[S4WRXTTCS]

I hope you americans soon can experience modern matrix lights. That is really amazing tech.

They're very close to being a reality in the US.

This Is Why Adaptive Headlights Are Now Legal In The US

The infrastructure bill, signed into law in November 2021, finally legalizes adaptive headlights.

www.hotcars.com

 

[VanFriscia]

I haven't read the article (paywall) but how is "Copilot" a better name than "Autopilot"?

I've heard this a few times that "Autopilot" should be called "Copilot" but to me that's backwards. By definition a co-pilot is a fully functional second pilot where you can pass off control and go to the bathroom, read a book, or take a nap. "Autopilot" is "a device for maintaining a set course" and does not allow a pilot to do the above (i.e. a pilot still needs to be engaged and aware). So it appears to me that "Autopilot" is the correct term to express these limits/expectations, not Copilot.

FSD should be called "Copilot" if anything (i.e. A Co-pilot is more capable than an auto-pilot, not the other way around). Also in FSD Beta and agree it's not full-self-driving nor Copilot capable, yet? .

You may be technically correct regarding the two terms, but I was looking at it with regards to how the average Joe would interpret them. I think most people would hear the "co" in copilot and think of it as an assistive feature. "Auto", however, implies autonomous - as in, kick back, relax and take a nap.

So what I'm saying is, if you don't know the literal definition of these terms, I think it's more likely that a person would interpret them backwards. Of course, I could be wrong about that. Just explaining my point

 

[EVNow]

So it appears to me that "Autopilot" is the correct term to express these limits/expectations, not Copilot.

The correct term would be "cruise control". Unfortunately that term has been already used for dumb CC, which should be called constant speed. Because of that marketing term denoting far greater functionality than actual functionality - subsequently all improvements are also exaggerated. So, you get TACC, then AutoPilot and then FSD. All of them are exaggerated marketing terms.

Like how LCD TVs are called "LED" or QLED etc.

 

[nvx1977]

What's flawed is the reliance on autobrights.

I have absolutely no issue with Pure Vision wanting to control brights. If it had control over turning on the bright I would simply observe when it did, and to make sure it was doing so only when necessary. But, it doesn't actually do that. It simply turns on autobrights, and gives no indication to the user whether brights are even necessary. I've gotten into the habit of always forcing autobrights back off after enabling FSD/AP.

I don't think FSD relies on headlights. I make it a habit of turning auto brights off every time I use FSD beta at night. It drives just fine. Often times better than day time. I'd imagine having a bit more visibility should increase FSD quality by some degree, but I doubt it is significant. This is why I believe Tesla gives us a way to turn off the auto brights. Like the 80mph limit on freeways, these restrictions should be relaxed as Tesla develops more confidence in FSD driving behavior.

In general, I never use high beams at night. I'm still blessed with good vision, and I'll just show down for blind hills/turns. I tend to think people overuse their brights. Tons of people here forget to turn them on when there's oncoming traffic. Can't wait for adaptive headlights.

 

[AndreP]

High beams are good for spotting wildlife, I see up to a dozen deer a day on my 20min commute to/from work

 

[apsen]

What's flawed is the reliance on autobrights.

I have absolutely no issue with Pure Vision wanting to control brights. If it had control over turning on the bright I would simply observe when it did, and to make sure it was doing so only when necessary. But, it doesn't actually do that. It simply turns on autobrights, and gives no indication to the user whether brights are even necessary. I've gotten into the habit of always forcing autobrights back off after enabling FSD/AP.

That's strange... For me it behaves similarly while on FSD beta or not. It controls them in both cases. I would say it is more often turns the brights off when unnecessary than the other way around. I.e. turning them off for the traffic signs, street lamps or lights around houses...

It's not really even about lidar.

That's the thing - it seemed like it was about lidar. Not sure how it could have been understood differently but ok, I'll chalk it up to English being my tertiary language...

 

[apsen]

It specifically said it was slowing the set speed due to weather. That was the first time I had ever seen any Tesla Adaptive cruise control system slow due to moderate rain. If it was heavy rain I wouldn't have been going that fast or if there was a lot of standing rain.

This is the first time I hear about something like that...

Anyone else sees behavior like that?

 

[apsen]

You may be technically correct regarding the two terms, but I was looking at it with regards to how the average Joe would interpret them. I think most people would hear the "co" in copilot and think of it as an assistive feature. "Auto", however, implies autonomous - as in, kick back, relax and take a nap.

So what I'm saying is, if you don't know the literal definition of these terms, I think it's more likely that a person would interpret them backwards. Of course, I could be wrong about that. Just explaining my point

I guess it could be like that if the person never heard of autopilot but I would expect that most people would have heard about it. I guess, I give an average Joe too much credit if that is not the case...