FSD Beta 10.13

[JulienW]

Had an easy five-second gap. Six second if you neglect an opportunity with traffic in a different lane.

Has to go in this situation.

Went with large truck on a five-second gap. If that is safe, the others are safe. Can’t have it both ways.

Feeling very good about my bet if that’s the margin they require. (The criteria I proposed fortunately clearly covered this case, as I knew it would be an issue.)

I thought it would be a close judgement call in some cases, but this one is not close.

By the way, it’s fine to not call this a failure, it’s just a failure by the clear criteria for that specific agreement. Caution is fine. Though I would prefer to not pull out and get passed by a massive truck.

(This was just for clip 3, I will look at the others now.)

Those cars look to be going >50MPH. 4 to 5 seconds is NOT enough time to safely accelerate to >50MPH.

 

[AlanSubie4Life]

Any bets should be based on whether Chuck Cook says that the car passes or fails the ULT.

That’s fine too. Just not what the beer is based on.

 

[Yelobird]

Any bets should be based on whether Chuck Cook says that the car passes or fails the ULT.

Agreed! Not based on shoulda, coulda, or I would have done it differently. From ”0” accomplishments prior to Safely pulling it off looks like a positive forward progression.

 

[jabloomf1230]

Agreed! Not based on shoulda, coulda, or I would have done it differently. From ”0” accomplishments prior to Safely pulling it off looks like a positive forward progression.

Oh, the clickbait opportunities there will be once Chuck has 10.13.

 

[Yelobird]

Oh, the clickbait opportunities there will be once Chuck has 10.13.

As there should be. This was likely not an easy computer task to pull off and he Positively pushed Tesla to solve the problem. Seems like they took his request more seriously then a .75 cent beer challenge with more rules then an HR manual. Hope he gets tons of hits for his valued contributions and efforts.

 

[AlanSubie4Life]

4 to 5 seconds is NOT enough time to accelerate to >50MPH except under full acceleration.

It doesn’t need to be that fast, because you are moving forward. You have more time than that, even if you want to leave a minimum 2-second gap.

In the case of leaving no gap (obviously cannot do that if the vehicle is in the same lane), you have 10 seconds, for a 5-second starting gap, assuming constant acceleration to traffic speed. (For that, your average speed is half of the traffic, so you’ll cover half the distance they do in that same time. So they will catch you at twice the gap time, 5 seconds after passing the turn (they have to go twice as far).) Not a problem. You have about 7-8 seconds to ensure the driver behind does not need to react. Very comfortable. Even my wife would not notice.

Also, again, if pulling out into traffic and having a truck pass you in an adjacent lane is ok, then other opportunities with more time and less risk need to be ok. The truck was not going substantially slower than other traffic. You cannot have it both ways.

1 for 4 is not bad though; I am getting a bit nervous since they could easily tweak the knobs and clear all but one failure.

 

[FloridaJohn]

That turn looked great. But, we don't know if, or when, FSD beta was engaged during the turn.

Yeah, for all we know, the driver could have gotten tired of waiting and took over to complete the turn.

We’ll have to wait for Chuck to get his chance before we really know how it’s performing.

 

[S4WRXTTCS]

I think the placement of this camera could be better to see what the traffic is like before the turn.

There could be things going on that we simply don't see.

The number one rule about turning left (or right) is to expect that someone will do something stupid.

 

[AlanSubie4Life]

The pictures you shared don't reflect how fast cars were approaching which is something a driver would consider and FSD may have as well.

The time stamps are really what I was capturing. You can see there is plenty of gap. And in fact it went on one of those gaps, after missing a couple.

Feels like your nitpicky here

Sure, but it's just following the conditions established in advance (which I think make it easier to judge): Elon: FSD Beta tweets

(I ended up agreeing to bet that it will have less than 90% success rate but removing condition 1 (which was dumb on my part, but presumably Chuck will test with traffic).)

How many times have you driven and thought to yourself you could have gone?

It happens. But this is not one of those cases since there's just one set of traffic to monitor and there's plenty of time.

There could be things going on that we simply don't see.

It's definitely possible someone was going really fast off camera and then slowed way down before entering the frame. I'm not sure what else we would be missing that would play into the go/no-go decision.

It did very well with the first crossing (and dealing with left turning traffic, which can be hard for a human and is a case you have to watch out for with this strategy!), which is a good sign. The second part is the easy part!

 

[Supcom]

Yeah, for all we know, the driver could have gotten tired of waiting and took over to complete the turn.

We’ll have to wait for Chuck to get his chance before we really know how it’s performing.

We also don't know what they were testing. It may be a candidate release of 10.13 It might be 10.12. It might even be a development version of 10.14 (or whatever comes after 10.13). Or something completely different created for this testing alone.

But, it's good to see Tesla putting effort into this UPL. I have similar turns in my area that should benefit.

 

[tmoz]

A few videos from today on Chuck's twitter, e.g.:

https://twitter.com/x/status/1551614099888431110

Frogger found that first safe log, but liked it so much it stayed put.

 

[AlanSubie4Life]

But, we don't know if, or when, FSD beta was engaged during the turn.

But, of course, the turn could have been all manually driven.

Yeah, for all we know, the driver could have gotten tired of waiting and took over to complete the turn.

I'm confident based on the behavior that FSD Beta was engaged. With the possible exception of when the car actually went (though it was probably Beta).

Frogger found that first safe log, but liked it so much it stayed put.

Haha, that's probably overstating it a bit. In the end it did go, and even if it was very conservative, it wasn't absurdly so. Even had a close call!

 

[Dewg]

We don't have a camera angle that shows the ramp-up traffic. We can't see if there are cars that were in the left lane, saw the Telsa waiting in the median, and changed lanes to the #2 lane, but the Tesla saw them in the left lane and didn't think it was safe. I never want a situation where my car has to "gun it" to get into a spot - like most humans would do. I've experienced this many times in Southern California traffic (known for a-hole drivers), where someone forces themselves into traffic, causing people to brake to let them in. Sometimes they have no choice, as traffic is so bad they'll never be able to complete the maneuver in a courteous manner, and end up stuck there for several minutes.

 

[AlanSubie4Life]

We don't have a camera angle that shows the ramp-up traffic. We can't see if there are cars that were in the left lane, saw the Telsa waiting in the median, and changed lanes to the #2 lane,

That's true, we could be missing that (though the gaps I pictured were for all three lanes so this wouldn't affect those). Fortunately when the actual 10.13 comes we won't have that uncertainty, though Chuck may have to add another camera angle or use the dash cam footage.

where someone forces themselves into traffic, causing people to brake to let them in.

Yeah, this is a failure per above criteria. But going briskly (0 to 50 in ~7 seconds) when someone is 6 seconds away will not result in that, and it's nice and comfortable easy driving.

As mentioned, Criteria #2 is definitely the hardest and the most nitpicking one. Because it's a balance, I said I'd just know it when I saw it, which obviously leads to a lot of latitude and room for disagreement. So these particular ones were failures to me (assuming nothing weird was happening off camera). But it was close. The main rationale and best argument for that is that it went with a closer gap, but didn't on the larger safer gaps.

 

[Dewg]

That's true, we could be missing that. Fortunately when the actual 10.13 comes we won't have that uncertainty, though Chuck may have to add another camera angle or use the dash cam footage.


Yeah, this is a failure per above criteria. But going briskly (0 to 50 in ~7 seconds) when someone is 6 seconds away will not result in that, and it's nice and comfortable easy driving.

If you ask any highway patrol or police officer, they will tell you that any maneuver that is safe is fine. I'd even argue that if the Tesla pulls out and causes the other car to cautiously slow down (let off the gas), it's fine. The other driver slowed down for their own personal driving experience - perhaps they are just a cautious driver, or know their reaction time isn't as it once was, so they slow down a bit. I do that myself from time to time. Someone pulls out in front of me, and I just let my foot off the gas a bit to keep the comfortable buffer between us. I don't know how fast their car is going to accelerate, and I don't want to assume. If the person has to brake, then it's not a safe maneuver.

 

[aronth5]

That's true, we could be missing that (though the gaps I pictured were for all three lanes so this wouldn't affect those). Fortunately when the actual 10.13 comes we won't have that uncertainty, though Chuck may have to add another camera angle or use the dash cam footage.


Yeah, this is a failure per above criteria. But going briskly (0 to 50 in ~7 seconds) when someone is 6 seconds away will not result in that, and it's nice and comfortable easy driving.

As mentioned, Criteria #2 is definitely the hardest and the most nitpicking one. Because it's a balance, I said I'd just know it when I saw it, which obviously leads to a lot of latitude and room for disagreement. So these particular ones were failures to me (assuming nothing weird was happening off camera). But it was close.

So on scale of 10 with 10 the highest how would you rate FSDbeta for Chuck's UPL based on what you've seen today compared to 10.12?

 

[Daniel in SD]

I’m feeling better about my bet but more pessimistic about Tesla’s approach.
If their simulation tools can't handle this scenario then what good are they? This seems like one of the easiest things to simulate.

 

[AlanSubie4Life]

So on scale of 10 with 10 the highest how would you rate FSDbeta for Chuck's UPL based on what you've seen today compared to 10.12?

I mean it's infinitely better than 10.12 according to this test. It could be successful 10% of the time and that would still be the case. So 10/10?

Overall this looks very promising, though, and it seems possible that I'll lose the bet. The positioning issues and potential false starts at the beginning of the turn could definitely give me the win though. Those are the things that define FSD, and I think it will be tough to suddenly fix them all (and it would be awesome if they were fixed).

 

[AlanSubie4Life]

If their simulation tools can't handle this scenario then what good are they

Maybe they're still in beta, not being used in production flow?

 

[Daniel in SD]

Maybe they're still in beta, not being used in production flow?

Looking at the AI day presentation they only talk about using simulations to provide data to train the perception neural nets so you could be right!