FSD rewrite will go out on Oct 20 to limited beta

[zecar]

I don’t think this will be too difficult. Meaning, there shouldn’t be a huge thinking process laborious enough to stall things, nor should there be a big risk of hitting someone. (In my opinion).

The computer doesn’t really have to determine intention. Just determine actions.

-If they are on the corner moving out, it waits/stops.
-If they are on the corner not moving, likely a brief pause, then go.
-If they suddenly start moving, the car stops again.
-If they are playing games moving in and out, or just moving around confusing the car, that could be a problem, but that’s not much different for us driving.
-At some point, we start moving, if they come out, we stop again.
Only the computer can do all this thinking quicker than us.

Waymo self-driving cars can now respond to traffic cops' hand signals - Roadshow

 

[M109Rider]

We need more videos. ...
I’m impatient.

 

[Mardak]

An unprotected left turn occurs at an intersection where there is no traffic light to signal the turn. The critical item is to understand that yielding is defined as "not causing any other drivers to change speed or direction due to your actions as a driver".

There's at least one from Tesla Raj's video although it doesn't look like there's actual oncoming traffic (but as you can see in the visualization, it briefly thinks there's something coming, so it yields):

 

[M109Rider]

There's at least one from Tesla Raj's video although it doesn't look like there's actual oncoming traffic (but as you can see in the visualization, it briefly thinks there's something coming, so it yields):

View attachment 601350
View attachment 601349

I saw that one. Very cool.

 

[drtimhill]

Will we move from beta to production in ten years?

... I said "early" beta! .. and yeah I remember how long Gmail was in "beta"!

 

[diplomat33]

Don't get me wrong, I am excited for the FSD Beta. It is great progress for Tesla. It is great to see FSD be able to handle intersections.

But that's a 70 miles trip with 4 interventions. It represents a disengagement rate of 1 intervention per 17 miles. That's not good. FSD is beta and it will get better but clearly it has a longs ways to go.

 

[2101Guy]

I hope that’s true. I’m really glad smart summon is in this. I love smart summon now, can’t wait for the new version !

i love it when
- it connects on the first attempt (rare, if ever)
-doesnt drive up and onto a curb when Im parked right next to the curb/parallell to the curb in the end spot, and the curb is between me and the car so that when it attempts to come to me, car doesnt pull forward far enough before turning and rear wheel goes up and ontop of the curb.
- it comes to me in a fairly direct manner down the parking lot vs hunting and feeling its way at 0.1mph

 

[Dutchie]

Why is everyone hanging on L2,L3,L4 etc. Yes, current system is L2 but only because Tesla and regulators said so. Perhaps it is technically way beyond that.

 

[Mardak]

But that's a 70 miles trip with 4 interventions. It represents a disengagement rate of 1 intervention per 17 miles. That's not good.

Have you driven that route in Sacramento to know what could be tricky? Has any other driving automation driven that route? Assuming no, that's 70 more miles than any other company. To be fair, unless there's someone in Tesla early access in Phoenix/Chandler, Waymo has done a lot more miles in that area than the FSD beta.

 

[qdeathstar]

Don't get me wrong, I am excited for the FSD Beta. It is great progress for Tesla. It is great to see FSD be able to handle intersections.

But that's a 70 miles trip with 4 interventions. It represents a disengagement rate of 1 intervention per 17 miles. That's not good. FSD is beta and it will get better but clearly it has a longs ways to go.

I think that is really good. There are a lot of decisions the car has to make. This represents a huge improvement over what is currently Public. Though before it gets to to public it will likely need to be much much better.

 

[qdeathstar]

Also, that’s 4 times where he “thought” he needed to intervene. It’s possible at least one of those times he didn’t actually need to.

 

[diplomat33]

Have you driven that route in Sacramento to know what could be tricky? Has any other driving automation driven that route? Assuming no, that's 70 more miles than any other company. To be fair, unless there's someone in Tesla early access in Phoenix/Chandler, Waymo has done a lot more miles in that area than the FSD beta.

I checked the route on Google Maps to get a sense of the type of route. But I do know that the disengagement rate is very low.

But I do look look forward to the video of the trip to see how FSD performed.

 

[Knightshade]

Why is everyone hanging on L2,L3,L4 etc. Yes, current system is L2 but only because Tesla and regulators said so. Perhaps it is technically way beyond that.

Given the only data we have is manual intervention needed 4 times on a 70 mile round trip, no, it's absolutely not way beyond L2.

Depending on the nature of the interventions, and how much warning was needed, it's POSSIBLE you could make an L3 argument- but I'd be very very surprised if there was enough warning (and a specific enough OD) to qualify as L3.



or to simplify:

It's not L3 or better unless it never requires immediate intervention inside it's OD. Not "rarely" but NEVER.

It's not L4 or better unless it never requires any intervention inside it's OD. Not '"rarely" and not "with 60 seconds notice" but NEVER.


It's not L2 today because "regulators say so" it's L2 today because it's not capable of higher level operation today.

 

[DanCar]

Why is everyone hanging on L2,L3,L4 etc. Yes, current system is L2 but only because Tesla and regulators said so. Perhaps it is technically way beyond that.

No, it is not way beyond that. There are places where there is no regulation prohibiting it, like Arizona. So it is just Tesla and will remain that way for years for L3 and beyond. Can't have a system that runs into curbs and other things. We will see such posts and videos as soon as this release has gone wide.

 

[diplomat33]

I think that is really good. There are a lot of decisions the car has to make. This represents a huge improvement over what is currently Public. Though before it gets to to public it will likely need to be much much better.

For comparison, Waymo has a disengagement rate in CA of 1 disengagement per 11,000 miles.

Yes, it will need to be much better.

 

[Mo City]

Don't get me wrong, I am excited for the FSD Beta. It is great progress for Tesla. It is great to see FSD be able to handle intersections.

But that's a 70 miles trip with 4 interventions. It represents a disengagement rate of 1 intervention per 17 miles. That's not good. FSD is beta and it will get better but clearly it has a longs ways to go.

I don't think you need to put up any more caution lights to pump the brakes. We all realize "there is a long way to go". Nobody thinks robotaxis are coming in 2 months (if ever).

It seems as if this is actually a "quantum leap" like Elon said.

Just enjoy the videos and the progress they indicate. It's sunny outside.

 

[powertoold]

For comparison, Waymo has a disengagement rate in CA of 1 disengagement per 11,000 miles.

Yes, it will need to be much better.

I remember when Tesla first released traffic light recog beta. It was around 98% accurate. We had seen some videos of it confusing one light with another, etc. A few months later, they released automatic stops at traffic controls, and it was a world of difference. So yes, it will get MUCH better.

 

[M109Rider]

i love it when
- it connects on the first attempt (rare, if ever)
-doesnt drive up and onto a curb when Im parked right next to the curb/parallell to the curb in the end spot, and the curb is between me and the car so that when it attempts to come to me, car doesnt pull forward far enough before turning and rear wheel goes up and ontop of the curb.
- it comes to me in a fairly direct manner down the parking lot vs hunting and feeling its way at 0.1mph

Ya, new technology is a tough nut to crack for sure.
That being said, I have no idea why companies like Tesla, Apple, Samsung, and LG, just don’t simply deliver a flawless result the first time, right out of the gates. ?

Why do we always have to go through this same old ridiculous process of improvements over time. Just deliver it perfect now right ??
Come on Tesla...

 

[raygduncan]

Clickbait from the Verge, the facts in the article mostly positive, but lots of gloom and doom from the writer (not all inappropriate but the general impression for the reader is "OMG are they crazy?!"

Tesla’s "Full Self-Driving" beta is here, and it looks scary as hell

 

[diplomat33]

I remember when Tesla first released traffic light recog beta. It was around 98% accurate. We had seen some videos of it confusing one light with another, etc. A few months later, they released automatic stops at traffic controls, and it was a world of difference. So yes, it will get MUCH better.

Yes, it will get a lot better over time.