AlanSubie4Life
Efficiency Obsessed Member
Mine works too.
It also works the other way.
Both can be true, just depends on the definition of “works.”
-
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
It also works the other way.
Both can be true, just depends on the definition of “works.”
Given the right set of circumstances, yes, the car can brake next to a truck or car or motorcycle. Given typical driving scenarios, with common circumstances, the car will not brake when next to those objects.
Neural networks can predict things based on common patterns of past "experiences" / training data. That's why we end up with false positives and phantom behaviors, but many times these are still reasonable, or when the thing exists, it's actually correct.
Yesterday I was waiting at a red light to go straight when the new high fidelity park assist activated rendering other vehicles and painted road markings such as lines and turn arrows. It correctly showed a left turn arrow for the left lane and right turn arrow for the right lane, but as the vehicle ahead moved for the green light, the visualization briefly rendered an imagined turn arrow under the lead vehicle before realizing there's no actual paint on the ground.
Even when vision is distorted by rain, it can predict things that are likely to happen based on common patterns of how things change, e.g., lead vehicle continues straight on the road. But the danger is predicting those when things have actually changed, so Tesla needs to spend extra effort in collecting and training with the many ways rain can affect the cameras of the many differing driving behaviors to reduce these mis-predictions if the goal is to not just disable FSD in heavy rain.
FSD will apply brakes if a vehicle (not just a large truck) veers toward the dividing line between lanes. It tends to happen mostly with large trucks since they are close to the lane lines to begin with and tend to have more issues tracking a straight line.
This is an area where the latency issues with V11 show up though, as the car often will apply the brakes after the truck has swung back away from your lane, giving things the appearance that the car is braking for no reason.
With the recent video processing improvements, I've seen less of this issue, but it still occurs.
AlanSubie4Life
Efficiency Obsessed Member
Keep in mind any level of occurrence that ends up being reported here is an extremely high rate relative to what is required for good overall end user experience.
This sort of thing really cannot happen. Our standards have to be very high for a functional L2 aid - even higher than for an L4/L5 aid, where people might accept compromises as long as it is safe enough for the manufacturer to take liability.
So let’s try to not minimize the complaint and claim your car “has the correct behavior” or “this does not apply to the entire fleet.” These statements are very likely not true in general, though it is possible (but subsequent post suggests not) you have never seen the issue.
This will also happen when FSD perceives this to have happened.
Given your logic, FSD will cause your car to crash. It's happened, and therefore will happen to everyone, given the right circumstances. But it's irresponsible to make bold claims like this.
For the majority of people with boring drives, the car will operate just fine.
S100D
Member
Thanks for the update on firmware and hardware.
My 2023 Model S LR was on 11.4.7.3 and I have hardware 4.0 when I had FSD disengage.
On 9-17-23, I was driving North on i-35 in Texas and I was in a very heavy thunderstorm. The speed limit was 75 mph and I was driving at 60 mph due to poor visibility. I received a message that due to poor visibility, FSD was degraded and the car slowed down 10 mph. I am 100% OK with FSD slowing down if there is too much water on the cameras. In another mile or so, FSD disengaged. If the visibility is very low, I am also OK with FSD disengaging.
What I am not OK with is that Elon has said many times that FSD can reach level 5 autonomy. He said this when I bought my 2016 S90D Model S with FSD. The difference is that my Model S90D had radar, ultrasonic sensors, and cameras. Unfortunately Tesla removed the radar and ultrasonic sensors to save money.
I am on 11.44.30.7 and I have not driven it again in heavy rain. I will respond again if I am in heavy raid and get any additional disengagements.
You don't think it's a fair statement that every single person has had a situation where FSD would or could have crashed if we didn't intervene?
Again this braking incident with trucks is commonly reported. From what I've seen, it's one of the most common complaints after incorrect lane selection into turning lanes both introduced with v11.
You act like FSD is near perfect, when most report the opposite.
Degraded and disengaging are two different things. On the 9-17 did your vehicle disengage?
Again, I'm not saying saying it didn't happen to you, but it seems like a rare case with v11. Most report it continuing to slow down and reporting FSD degraded, but staying active, regardless of weather.
V10 would revert to AP and sometimes would even disengage. It's obviously something Tesla has worked on.
With FSD, I believe all anecdotal stories. The weather, environment, situation, and even individual car create different experiences. It's still pretty far away from consistency, but I see what Dewg is saying with the most common behaviors across the fleet.
Although, it appears that more drivers than not have experienced poor/aggressive braking due to trucks.
I think we all can agree there's a lot of work to do to get consistent level 2 and more for 4/5. We are still years away from that.
Although, it appears that more drivers than not have experienced poor/aggressive braking due to trucks.
I think we all can agree there's a lot of work to do to get consistent level 2 and more for 4/5. We are still years away from that.
Can we direct all reports on current behavior of FSD to the FSD v11 thread? I'd like to read and talk about v12 in this.
AlanSubie4Life
Efficiency Obsessed Member
I believe that logic to be correct. I also think that people are really quite good at monitoring particularly in risky situations, and in addition the new driver monitoring system works pretty well, and users will tend to intervene to prevent crashing. So in spite of FSD's best efforts, it probably rarely happens and definitely won't happen to everyone.
I've not had an event for a while where I really thought the car would have crashed without me intervening, but I often intervene early, and there's so much marginal behavior that I immediately terminate, that it's hard for me to say for sure that there weren't any dangerous situations which would have evolved into a crash (probably due to someone running into me from behind).
But anyway that doesn't make me conclude that it can't cause my car to crash. I totally believe it could run into a fixed object given the exact right scenario. Just never seen it.
Last edited:
Actually you are the one incorrect. On my last road trip, maybe 3 months ago, on whatever the current FSDb version was, on a 2-lane highway I'd be right about at the middle of the tractor-trailer when I got a serious brake. Absolutely horrible. I have my own way of passing trucks, a habit from when stuff that kicks up would ding my paint (not currently a problem) where the key is to overtake smoothly, so that sudden brake for no friggin' reason got real old real fast. After 4 or 5 times, I'd just disengage to get past the trucks.
So it never occurred at all in my first 20K miles or so of road trips, beginning Spring '22. On my last one, it would happen every few trucks, and none of them ever stinkin' swerved. Not at all.
I'd love to read reports about V12 in the field. I think so far in this space-time continuum hasn't there been... one? lol.
I was describing the correct behavior. You are describing your experience, and then saying my description of the correct behavior is incorrect. And then you say that your car didn't do that for the first 20k miles, and now it is doing it constantly. So are you staying the correct behavior now is to slam on the brakes? Or are you saying your car is slamming on the brakes and something is wrong with your car?
Yeah, so let's speculate about v12 capabilities and its potential wide release timeline. v11 stuff belongs to the v11 thread.
S100D
Member
Yes, I got a message that said due to poor weather, FSD was degraded and this is when my Model S slowed down 10 mph. A few miles later, FSD disengaged (turned off) and Enhanced Autopilot took over, Enhanced Auto Pilot also disengaged due to the bad thunderstorm a few miles later. If the cameras are so wet that they cannot provide accurate data to Full Self Driving, FSD/Auto Pilot should disengage. I have to admit, it was one of the worst thunderstorms I have been in, in a long time. I was only in this heavy rain for about 15 miles.
I have also gotten messages with FSD turned on when it was sunny and cloudy (no rain) that said "Poor weather, FSD degraded". The car and cameras were clean. It is unbelievable that when the sun is shining on one occasion and there are clouds on a different day, FSD thinks the weather is poor.
Similar threads
