MadOverlord
Member
A friend posted this photo on FB, it's in San Francisco. I'd dearly love to see the FSD beta do a u-turn from the second from the left lane.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD rewrite will go out on Oct 20 to limited beta
- Thread starter diplomat33
- Start date
MarcG
Active Member
A friend posted this photo on FB, it's in San Francisco. I'd dearly love to see the FSD beta do a u-turn from the second from the left lane.
Yep, this is the intersection of Mission with Van Ness in the northbound direction.
Many humans get confused themselves, so I wouldn’t be surprised if FSD messes it up.
powertoold
Active Member
So at best it goes to EAP at v.15, but likely v.20 plus.
Then another 8-18 mos before general release. This is smart summon all over again, and we all know what the end product was...
Last edited:
powertoold
Active Member
FSD beta is actually mind blowing and impressive though. Tesla is making very fast progress towards "level 5" robotaxis, which many people didn't think was possible until 2030.
They’re on rev 8 right now? There better be some major improvements in the next couple then.
Years ago, when the tech was more primitive, yeah, many people did not see L5 happening before 2030. But as the tech has gotten better, L5 looks to happen sooner. Today, I think most AV companies say L5 will happen before 2030. For example, Mobileye is saying they will deliver L5 as ride-hailing by 2022 and L5 in consumer cars by 2025.
DanCar
Active Member
Unlikely level 5 will happen before 2030. I'll by happy if I'm wrong. Although level 4 that works in 99.9+% of the situations will be very helpful. It will be quite a challenge to handle snow, ice, flooded roads, fog, dust storms, poorly designed intersections, harassing drivers, teenagers getting in the way with a death wish, signs based on potential weather conditions, etc...
Last edited:
Knightshade
Well-Known Member
L5 is not happening with the current sensor suite.
Write it down- then re-read it every new years eve as it keeps being correct forever.
Even NoA still turns off in only moderate rain because other than forward there's only 1 camera covering most of the way around the car and with no way to clear rain.
The current cameras have blind spots (particularly close and low to the vehicle- and to the sides if any object is blocking one of the side cameras) too.
I certainly think something like L3 highway is very doable with current sensors... MAYBE L4 with some significant ODD bounding. L5? Not so much.
I think L5 they're gonna need to add some more cameras both for redundancy and simply to see the stuff they can't see in many everyday situations... (something like both regular and fisheye lenses on the lower front and rear fenders would allow both a "real" overhead 360 parking view, and easy cross-traffic visiblity when say the B-pillar cameras can't see well from an object... and honestly a rear radar wouldn't be the worst idea either... nor would cameras capable of higher framerates if they're gonna rely on just visual for speed estimation at distance).
Right, the Early Access Program that they lied to early FSD customers about getting invites for.
It'll be a long while before they release it to us peasants who aren't social media influencers or major investors.
Tiger
Active Member
Tesla should already release the new AI to everyone so that it can already start learning (and improve) from Autopilot scenarios. Does not make sense wait for full FSD or FSD beta, just restrict it to Autopilot.
Doesn’t work like that. All of the driving instructions, afaik, are hand coded. NNs handle perception only. How to deal with what it sees relies on humans programming the instructions. Until Dojo is up, there’s no NN handling driving instructions. More data than what they’re getting already won’t improve anything more than how much they’re already improving. If more data would help Tesla, you could be sure they’d be getting more already.
Tiger
Active Member
Sounds like spaghetti to me, maybe few more rewrites?
Anyhow, it should be possible to upgrade Autopilot to drive as well as the FSD Beta (just without navigation).
Again, that's not how the software/NN stack works. The current autopilot uses an older NN stack which has not been changed. The new stack is (initially) focused on solving problems related to city driving (though a large part of it can be applied to highway driving also at a later time). And, as others have noted, Tesla have enrolled enough FSD beta drivers to give them all the data and feedback they can cope with .. more would just be duplication. As the stack matures based on this feedback they will be able to open it up to a larger set of testers to focus on edge cases.
Tiger
Active Member
Dirty Tesla has shown it works better also in non city situations, they should already roll out these benefits to Autopilot users. It's a severe design flaw if they can't leverage feedback from a wider Autopilot fleet. They should be able to expand gradually and also get feedback gradually, starting from simpler scenarios and expanding to more elaborate scenarios.
Last edited:
How would ou feel about them rolling it out with confirmations required for various maneuvers that are deemed risky?
Similar threads
- Question
