Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD Beta 10.13

This site may earn commission on affiliate links.
I can't imagine accidents not happening at some time soon with the increased numbers of beta testers. Not that the new people are less careful, but that the number of times the dice are getting rolled is increasing. Then there will be consequences...
That's what you would think based on the collective evidence so far, but I venture to offer a different perspective on the future "trend". Some of the reasons that contribute to FSDb's unpredictable behaviors is the fact that it's trying to perform in the realm of human drivers, which by definition are unpredictable. Imagine, when we get to a point where all driving is done by some FSD-equivalent algorithm (i.e. autonomous), then every scenario is predictable. You'll not have someone jump a 4-way stop out of turn, or make a prohibited right turn on red. In that eventual state, "confidence" will no longer be an issue. So, I see this trend as becoming easier and easier with more adoption of autonomous driving, not harder. There may be a temporary increase here and there due to not understanding the nature, premise, and implications of FSDb, but those will root themselves out over time.
 
I don't think FSD will have that many accidents unless it's thin or short objects the camera can't see. FSD is very slow. Also, I can see why some people love FSD today.

I took my car to Tesla service and was given Uber credits. This 5 star driver was like FSD. He went the speed limited at 45, but people do 60-65 here. He would slow to 10 mph to make a turn, but slows down to 5 before completing the turn. He was consistently making large adjustments to speed and wheel turn rate at the same time. Yes, he was in unfamiliar area, so maybe that's why, but he reminded me so much of FSD that I realize maybe many people drive like him so that's why many people think FSD is really good.

When I drive I commit, I plan to put my tires where I want before the turn, I know my speed in and speed out ahead of time. I make 1 steering wheel motion in and 1 smooth motion out. I used all the road in my lane. Maybe FSD is just not for me.

Anyway, will test 10.13 when I get it but as usual I keep it off after a short trip ot 2 as it does very dumb things like spin my wheels left and right as it tries to make a turn and sometimes stops mid turn for no reason before continuing.
 
I don't think FSD will have that many accidents unless it's thin or short objects the camera can't see. FSD is very slow. Also, I can see why some people love FSD today.

I took my car to Tesla service and was given Uber credits. This 5 star driver was like FSD. He went the speed limited at 45, but people do 60-65 here. He would slow to 10 mph to make a turn, but slows down to 5 before completing the turn. He was consistently making large adjustments to speed and wheel turn rate at the same time. Yes, he was in unfamiliar area, so maybe that's why, but he reminded me so much of FSD that I realize maybe many people drive like him so that's why many people think FSD is really good.

When I drive I commit, I plan to put my tires where I want before the turn, I know my speed in and speed out ahead of time. I make 1 steering wheel motion in and 1 smooth motion out. I used all the road in my lane. Maybe FSD is just not for me.

Anyway, will test 10.13 when I get it but as usual I keep it off after a short trip ot 2 as it does very dumb things like spin my wheels left and right as it tries to make a turn and sometimes stops mid turn for no reason before continuing.

I remember driving to Springfield, MA from Boston for a business trip. Driving was slow and courteous out west. Reminded me of my childhood roads in suburban Atlanta. Couldn't call anyone in Springfield a Masshole. FSDb might do really well there.
 
  • Like
Reactions: sperkin
I remember driving to Springfield, MA from Boston for a business trip. Driving was slow and courteous out west. Reminded me of my childhood roads in suburban Atlanta. Couldn't call anyone in Springfield a Masshole. FSDb might do really well there.
Not in LA, but I might just need to be more patient and realize it's not going to do what I have planned in my mind.

FSD was actually good at 3am except handling 1 merging car into my lane on the freeway, but got me home fine. Just normal usage in traffic it's really really slow. Every turn it seems like it's calculating and making decisions every millisecond. It's not confident and commit. It also picks the wrong lane to be in or go into the wrong lane at the wrong times.
 
Every turn it seems like it's calculating and making decisions every millisecond. It's not confident and commit.
Maybe not every millisecond, but every 50-100 milliseconds? Absolutely.

Human high level reasoning and planning and decision making is extremely slow relative to subconscious reaction and reflex. Right now FSD is replanning and optimizing and reacting at superhuman speeds. I'm not sure how much of my discomfort with its behavior is because those multiple tiny corrections are consistent with a young driver who hasn't yet learned how to handle complex maneuvers, so relies on fast and jerky corrections as the maneuver evolves.

Tesla's overriding priority right now is safety. They know that if a fatal crash occurs there will be a public furor and multiple proctoscopic/cystoscopic investigations, and in retrospect any code choices or neural net weighting towards comfort over safety could have a catastrophic impact on the program and the company.
 
  • Like
Reactions: BnGrm
Maybe not [making decisions] every millisecond, but every 50-100 milliseconds? Absolutely.
Some of the things that were going into the new, integrated vision stack (which I understand will be in v11) were 1) moving from "bag of points" to vector space before driving NNs, 2) moving from separate 8 camera vision processing to vision processing of one video feed that was 8 cameras stitched together, and 3) adding levels of recurrency in the vision processing and classification NNs. I wonder where 10.13 stands on the progression to these items - especially #3 where recurrent NNs should have a lot less "jerkiness" in decision making and be smoother (recurrency takes the predictions from some number of past calculations and feeds them into the calculations for the current prediction).
 
Can anyone tell me what this (see atached jpg) is for?
1655849228965.png

It used to be used on NOA on the freeway. Now the single blue 'guide line' is already present on streets. It will change it still on the freeway, but I think it's still self driving without it?
 
Can anyone tell me what this (see atached jpg) is for?
View attachment 819360
It used to be used on NOA on the freeway. Now the single blue 'guide line' is already present on streets. It will change it still on the freeway, but I think it's still self driving without it?
It is still NoA. FSD Beta doesn't work on freeways. You are still using the same NoA Stack when you are on a freeway.
 
I took my car to Tesla service and was given Uber credits. This 5 star driver was like FSD. He went the speed limited at 45, but people do 60-65 here. He would slow to 10 mph to make a turn, but slows down to 5 before completing the turn. He was consistently making large adjustments to speed and wheel turn rate at the same time. Yes, he was in unfamiliar area, so maybe that's why, but he reminded me so much of FSD that I realize maybe many people drive like him so that's why many people think FSD is really good.
Or maybe they just remember early versions of Autopilot on HW2/2.5 on CA-17, back when a slightly inebriated student driver would be better, so FSD looks particularly good in comparison. :D
 
It's going to be v14.0.2 that's the gamechanger. Nothing will work properly until that version is released, so you should just opt out of Beta and disable AP/NoA until then. Save yourself some stress. 😋
We have been on 10.x for ¾ of a year and no end (11.x/single Stack) in sight so 14.x should be in about 2025. And that is probably Elon time which could put it closer to 2030.