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

FSD Beta Videos (and questions for FSD Beta drivers)

This site may earn commission on affiliate links.
That ride was epic fail. I lost count how many disengagements and near collisions it had.

With this video and the Seattle video trying to hit monorail pillars and planters I would say that this build was seriously overhyped.

Yes, there are some incremental improvements and eye candy, but the most basic functionality, i.e. “do not hit objects“ still lacks.

What, you're saying that it didn't blow your mind?
 
I'll save you some time. No, Waymo has not been seen to make or attempt Chuck-level unprotected lefts.
Yeah, luckily the JJ page is annotated as diplomat mentioned. There are some unprotected lefts from stop signs but on two lane residential streets with no cross traffic. It would be interesting to see how the various technologies would handle a Chuck level unprotected left.
I'm impressed by Waymo and FSD 9 so far and am glad there are so many competing approaches trying to solve the problem. The Downtown SF video is frustrating. It almost seems like there's something wrong with his Tesla though... those left turns were bizarre and pretty unique to that video.
 
The Downtown SF video is frustrating. It almost seems like there's something wrong with his Tesla though... those left turns were bizarre and pretty unique to that video.

I don't think so. We've also seen major 8.2 fails in downtown Chicago, in similar fashion. V9 is definitely better, as the video guy says, but SF has situations that are particularly challenging for FSD beta, mostly lane choice and lane changing / skipping ability. I drive in SF all the time, and it's funny to see fsd beta get confused about the same things as me (e.g. V9 taking the right lane over the bridge, because it seems like both are going straight, near the beginning).

Also, Fsd beta often gets confused turning into multi-lane one-way roads. We saw this a lot in downtown Chicago.

Edit: I wouldn't blame fsd beta fails on "something wrong with the Tesla." I made this mistake initially with brandone's videos a long time ago, but it turns out fsd beta just sucks in certain environments. Later on, brandone's fsd got dramatically better, so the whole Tesla problem theory got debunked. I don't think this will be the case for SF any time soon though.
 
Last edited:
E6AEGA_VcA03CWb
 
#FSDBeta 9.0 - 2021.4.18.12 - Drive with some interesting lane changes, and a Smart Summon Test

Routing/mapping looks terrible in Jacksonville. The car was making some weird decisions on lane changes, the worst of which looked like the car trying to turn right when the GPS clearly said it needed to keep going straight for a few more miles before reaching a LEFT turn.
 
  • Like
Reactions: Matias
I don't think so. We've also seen major 8.2 fails in downtown Chicago, in similar fashion. V9 is definitely better, as the video guy says, but SF has situations that are particularly challenging for FSD beta, mostly lane choice and lane changing / skipping ability. I drive in SF all the time, and it's funny to see fsd beta get confused about the same things as me (e.g. V9 taking the right lane over the bridge, because it seems like both are going straight, near the beginning).

Also, Fsd beta often gets confused turning into multi-lane one-way roads. We saw this a lot in downtown Chicago.

Edit: I wouldn't blame fsd beta fails on "something wrong with the Tesla." I made this mistake initially with brandone's videos a long time ago, but it turns out fsd beta just sucks in certain environments. Later on, brandone's fsd got dramatically better, so the whole Tesla problem theory got debunked. I don't think this will be the case for SF any time soon though.
I meant those left turns were unusual compared to the other FSD v9 videos I've watched. The two that happen after the five minute mark are weird. There were a couple of wide left turns on one of the Chuck videos but nothing as alarming as the SF video.
But yeah, a lot of the issues seemed to be lane related and I was equally confused!

Overall though, I'm very impressed with the videos of the first iteration of v9. They obviously have work to do but some of those turns are looking very smooth and natural.
 
Routing/mapping looks terrible in Jacksonville. The car was making some weird decisions on lane changes, the worst of which looked like the car trying to turn right when the GPS clearly said it needed to keep going straight for a few more miles before reaching a LEFT turn.

I honestly thing tesla is further ahead on vision then they are on mapping/driving policy. Almost all of my disengagements on NoA are because either their map is wrong or it interprets the map wrong. Might almost be harder to get right then anything else in self driving. I haven't ever heard tesla (elon or karpathy) talk about this challenge, would be interesting to know what their plan is.
 
I honestly thing tesla is further ahead on vision then they are on mapping/driving policy. Almost all of my disengagements on NoA are because either their map is wrong or it interprets the map wrong. Might almost be harder to get right then anything else in self driving. I haven't ever heard tesla (elon or karpathy) talk about this challenge, would be interesting to know what their plan is.

Yeah, AP has always picked the wrong exist lane on a route a take quite often. If you take that lane, you end up "stuck" because the lane divider immediately becomes a solid white. I wish there was a way to let Tesla know about these nav issues. I've tried letting AP get in the wrong lane and then disengaging but that never seems to get anyone's attention.
 
Agreed. Next FSD beta update, unprotected left videos from Chuck is all I am interested in

Anybody want to place bets on whether they end up retrofitting two new side-facing cameras into the front bumper, reusing the no-longer-relevant RADAR wiring? :)

SF has situations that are particularly challenging for FSD beta, mostly lane choice and lane changing / skipping ability. I drive in SF all the time, and it's funny to see fsd beta get confused about the same things as me (e.g. V9 taking the right lane over the bridge, because it seems like both are going straight, near the beginning).

This is why I won't drive in SF unless I absolutely have to. I can't even imagine beta testing FSD in SF. About the only worse idea I can think of would be taking your driving test there. :D
 
Anybody want to place bets on whether they end up retrofitting two new side-facing cameras into the front bumper, reusing the no-longer-relevant RADAR wiring? :)
...
I wont bet, but to me it's the most glaring deficiency in the FSD sensor suite. A basic angle-of-view limitation that is evidently confirmed by all the excessive creeping-forward behavior, and something that you'd think should have been very obvious in the early platform design. However I hate second- guessing other clearly capable designers, so I reserve some judgment to be proven wrong - which hasn't happened yet from the YouTube evidence.

Car-platform-wise, my suggestion would be to fit side-looking cameras into new custom headlight housings (and I've also posted on sensible ideas for camera-friendly headlight improvements that can help camera-Vision performance, without the all-or-nothing issues of existing auto high beams).

This idea would be compatible with field retrofits of older cars. I said compatible not super-cheap or DIY-simple.

However, I think a limitation is the number and protocol of sensor connectors in the HW3 computer. IDK but it seems unlikely that the radar data port can be repurposed in-situ for additional camera(s). Possibly a video pre-processor module that blends say the left & right windshield wide-view camera feeds with the new left & right corner camera feeds, so that the stream into the existing camera ports can use it without a complete redo of the FSD hardware configuration.
 

just started watching this vid, but wanted to say that the first maneuver here, an unprotected left that is fairly simple, is very well done. Car moves forward on a green a little across the crosswalk, waits for the oncoming car and two bikers in the perpendicular crosswalk to pass with a STRAIGHT STEERING WHEEL (!!! Finally!!!), and then finally goes when it’s clear. Closest I’ve seen so far of proper behavior on an unprotected left at a light. More proper would be going further into the street and not twitching the wheel so much and moving more assertively once clear, but overall nice.