-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD v12.x (end to end AI)
- Thread starter Buckminster
- Start date
-
- Tags
- #fixphantombraking autopilot chill is not chill elon is an ass fix fsd extreme late deceleration fsd does not commit to go fsd does not maintain proper speed fsd hard and late braking fsd jerky driving fsd late deceleration fsd stops too early too quickly fsd stops too far behind line fsd tailgating in city streets
Tesla Vehicle Safety Report says "That’s why Tesla vehicles are engineered to be the safest cars in the world." But looks like it's been relatively flat for 2023 (especially if you back out 2022 Q1's jump that gets dropped off in 2023 Q1 in my trailing chart):
It would make sense to use FSD technology to increase safety of all Tesla vehicles as standard Autopilot when it's ready. Making it "free" or included might upset those who paid, or Tesla could figure out ways to continue to provide exclusive benefits.
sleepydoc
Well-Known Member
4 way stops are intrinsically difficult, even for humans. How do you train a neural net with examples when humans can’t get it right, either?Agree BUT he is the "only game in town".
One of my peeves is 4-Way Stops. V12 is just as horrible as V11. It happens at the very first intersection. V12 sits at the Stop Sign and allows 4 cars to go ahead of it. It is so inept at taking the initiative at 4-Ways and just sits there allowing everyone to take advantage of it. I never (unless there is no other traffic) get through a 4-Way without accelerator intervention.
Yeah, this might be easier than I originally thought. It might be relatively straightforward to have Basic Autopilot reuse FSD except drive as if there's no navigation destination with Minimize Lane Changes (assuming this still works with end-to-end control). This should allow it to stay in lane and not choose to make turns at intersections unless in a lane forced to turn. So this should increase the safety of being able to go around parked vehicles and giving distance to vulnerable road users, etc.
All the other people not using Autopilot have to hit something...
powertoold
Active Member
Tesla AI probably released v12 to Omar early so we can gauge how fast the progress is
And I know people here like Chuck and Dirty Tesla et al, but Omar does a lot more variety of routes than any of the other influencers
I get that we want to see Chuck's left, but his drives are pretty dull otherwise
And I know people here like Chuck and Dirty Tesla et al, but Omar does a lot more variety of routes than any of the other influencers
I get that we want to see Chuck's left, but his drives are pretty dull otherwise
Repeated drives are good for regression testing. I'd much rather see Chucks route than some cherry picked Whole Mars route.
I'd rather see my own frequent routes.
powertoold
Active Member
When I watch Omar videos, I already know what it's gonna do on Chuck's left: freeze
Omar's drives have almost everything I want to see except for heavy rain (already know V12 isn't good here) and snow
jebinc
Well-Known Member
I
I think “Black Tesla” does the most objective reviews. Always uses the same route for comparative purposes as well.
AlanSubie4Life
Efficiency Obsessed Member
Yeah. Unfortunately we have no data so it's impossible to say what is contributing to the plateau.All the other people not using Autopilot have to hit something...
It's unfortunate that there is no data on safety.
While the naysayers continue to bash Omar I thought this exchange with AI Driver was telling and overall there were many good exchanges between them.
@24:05 Omar talks about V12 being the biggest update ever and AI Driver says "that is certain without a doubt".
@24:05 Omar talks about V12 being the biggest update ever and AI Driver says "that is certain without a doubt".
Last edited:
sleepydoc
Well-Known Member
Hard to tell for sure and we don't know exactly what the data means. In the report, they say "We collect the amount of miles traveled by each vehicle with Autopilot active or in manual driving" but they don't specify what 'autopilot' means. Is that just AP and EAP or does it include FSD as well? Or has it changed since they combined the AP and FSD stacks?
it's hard to say for sure if there's a plateau or not. They may be but it's also important to remember that as the accident rate gets lower and lower it gets harder to achieve gains. There will always be random events that they can't anticipate or control for. As the preventable events get fewer and fewer those random events make a larger proportion of the accidents making further gains increasingly difficult.
Todd Burch
14-Year Member
I am sure the autopilot team is watching Omar's videos. I am certain they are aware of the excess hesitancy at 4-way stops and are working on addressing it.
I imagine that every three weeks or so, Tesla can train a new model based on insight gathered from use of the previous model. Remember, iteration should be faster now because we're not talking about writing bug-prone C++ code to handle various situations. We're using the data engine now--test a network, determine the weak spots, source data addressing those weak spots, & retrain with that data. And as Tesla's compute grows this year, that turnaround time will get even faster because Tesla's ability to train the network will get faster.
I imagine that the version of v12 that we'll get in a few weeks will be at least one iteration of additional training compared to the network that Omar's driving on now. It will be fascinating to see how the pace of progress changes. From everything I know about neural networks, the pace of improvement should get faster now, not slower. I would imagine we'll see FSD version updates much faster compared to v11. Like maybe an update every 2 months. Once they get into a rhythm of training and testing a new iteration as the previous iteration is downloaded to our cars, I feel like we'll see steady progress now.
Omar mentioned some others might be getting v12 this weekend. This is an exciting time.
For the longest time we've seen a lot of negativity in these FSD threads. And I get it. We all want it now, people feel frustrated, etc.
But just keep in mind that this is at the very forefront of human technology. Try to be patient and appreciative of the amazing things we're seeing come out of this system, as imperfect as it currently is.
I imagine that every three weeks or so, Tesla can train a new model based on insight gathered from use of the previous model. Remember, iteration should be faster now because we're not talking about writing bug-prone C++ code to handle various situations. We're using the data engine now--test a network, determine the weak spots, source data addressing those weak spots, & retrain with that data. And as Tesla's compute grows this year, that turnaround time will get even faster because Tesla's ability to train the network will get faster.
I imagine that the version of v12 that we'll get in a few weeks will be at least one iteration of additional training compared to the network that Omar's driving on now. It will be fascinating to see how the pace of progress changes. From everything I know about neural networks, the pace of improvement should get faster now, not slower. I would imagine we'll see FSD version updates much faster compared to v11. Like maybe an update every 2 months. Once they get into a rhythm of training and testing a new iteration as the previous iteration is downloaded to our cars, I feel like we'll see steady progress now.
Omar mentioned some others might be getting v12 this weekend. This is an exciting time.
For the longest time we've seen a lot of negativity in these FSD threads. And I get it. We all want it now, people feel frustrated, etc.
But just keep in mind that this is at the very forefront of human technology. Try to be patient and appreciative of the amazing things we're seeing come out of this system, as imperfect as it currently is.
zoomer0056
Active Member
Don't use Darwin award winners for training. It's not that difficult if you know the rule that the person on the left has to yield. If there's a four way tie situation then one can inch forward to judge the others reaction and safely act accordingly. There are rules, however some don't know them and others ignore them. So always yield to the unsafe driver if it calls for that.
Last edited:
AlanSubie4Life
Efficiency Obsessed Member
I’ve always recognizedi it will take years and I recognize that it will take years with v12 to get to a useful L2 City Streets feature that is more relaxing and safer than just driving oneself.
What makes me negative is people who underestimate the task ahead and claim we’re near autonomy. That just leads to disappointment.
Yes, but the only way of teaching FSD about stop signs is to give it a million examples* of people negotiating at stop signs. If we consider a million examples* of "average drivers" at stop signs, I can imagine that the system wouldn't actually have any idea how to proceed when other cars are around. Tesla may not have gotten around to carefully curating the examples. They may still be high fiving each other over the behavior when no cars are around.
*A manufactured number. I have no idea how large the data set is.
zoomer0056
Active Member
Stick to the letter of the law except to yield for safety. The car absolutely knows the law but must always take action for safety. It's knowing how to react to avoid an accident. But I really don't know how they're implementing any of these decisions.
diplomat33
Average guy who loves autonomous vehicles
Why does the Tesla in the thumbnail graphic have what is clearly a Waymo-like sensor pod on the roof? What's that about? I guess they are trying to make the Tesla look like a robotaxi. LOL.
Similar threads
- Article
