-
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
If every other of those disengagements would have lead to an accident (if car was driving autonomously), the current rate would be one accident every 40 miles. If driver normally is involved in an accident every 400 000 miles, that distance needs to be 10 000 times higher.
So if disengagement doubles every year it would reach 400 000 in approximately 13 years. Seems plausible. If the actual accident rate now autonomously would be 1 in 100 miles, then we would reach parity already in 12 years!
Last edited:
Pdubs
Ingesting large quantities of LSD
Yeah man. This happens at intersections for me as well when there’s not enough room to clear it.
sleepydoc
Well-Known Member
Agree, but the one advantage FSD does have are the front cameras are higher on the windshield so the cameras have a better view then humans and will see the pothole sooner. Whether that slight advantage will actually mean anything I don't know but I'm doubtful.
Perhaps part of the solution is to communicate to the fleet. Even avoiding the slightly depressed sewer covers in my area would be helpful since I typically avoid them as well.
EVNow
Well-Known Member
An L4+ accident happens every 10k miles - simple things like curbing wheels. You can check my older posts for links.
I don't think every other disengagement would prevent an accident - where as some interventions may have prevented accidents (say going across the intersection slowly so you push the accelerator). My disengagements or interventions were very rarely to prevent accidents. Only Tesla knows the real numbers (if at all).
If we just go by the crowdsourced numbers, may be one in 10 disengagements would be to prevent to accidents.
I had a similar experience the other day when a car wanted to pull out from a CVS parking across 2 lanes to turn left. FSD stopped short of the line of cars in front of me and let the car cross to make the turn. At first I wasn't sure why FSD stopped until I saw the car trying to come out. I was shocked but in a good way.
V12 also consistently avoids pulling into intersections at lights to avoid blocking traffic in case the light turns red.
Do you mean than very minor accidents for humans happen every 10 k miles?An L4+ accident happens every 10k miles - simple things like curbing wheels. You can check my older posts for links.
I don't think every other disengagement would prevent an accident - where as some interventions may have prevented accidents (say going across the intersection slowly so you push the accelerator). My disengagements or interventions were very rarely to prevent accidents. Only Tesla knows the real numbers (if at all).
If we just go by the crowdsourced numbers, may be one in 10 disengagements would be to prevent to accidents.
View attachment 1046457
A pilot friend shared one with me that I now cannot get out of my head any time I hear someone misuse the word: when the flight attendant announces "we'll be landing momentarily", I picture the pilot doing a touch-and-go...
My job as a driver is to ensure my safety, not worry about impatient drivers running over to get a quickie with their girlfriend
I'm not sure if I treat FSDS differently from others, but I have two distinct driving styles.
First is experiment/lab where I will give FSD time to figure things out and make mistakes. Usually there is nobody else in the car to witness this.
There are way fewer disengagements but many more mistakes and odd behavior.
The Second is mostly when I have passengers/family etc in the car. Then I don't even engage FSD if I think its going to mess up, or that its messed up in this bit of road. Significantly more disengagements for this type of drive because the slightest hint of a mistake causes me to disengage (single pull for the win).
Most of the things I report here are from the first type of drive. Although sometimes FSD does something bad in the places it has previously done well.
But it's the second type where the family is never sure if it's me or FSD in control. This is where I think FSD does really well because it is much closer to driving like a human.
Can't wait to see what gets added into 12.4/5
First is experiment/lab where I will give FSD time to figure things out and make mistakes. Usually there is nobody else in the car to witness this.
There are way fewer disengagements but many more mistakes and odd behavior.
The Second is mostly when I have passengers/family etc in the car. Then I don't even engage FSD if I think its going to mess up, or that its messed up in this bit of road. Significantly more disengagements for this type of drive because the slightest hint of a mistake causes me to disengage (single pull for the win).
Most of the things I report here are from the first type of drive. Although sometimes FSD does something bad in the places it has previously done well.
But it's the second type where the family is never sure if it's me or FSD in control. This is where I think FSD does really well because it is much closer to driving like a human.
Can't wait to see what gets added into 12.4/5
EVNow
Well-Known Member
Yes.
Autonomous Car Progress
I think supervised FSD is a terrible idea, and is going to fall on its face so I agree with you on that one. I don't doubt that some version of FSD (or a competitors version) will be safer than the average human driver, but I don't think that's a good comparison. The entire comparison to...
teslamotorsclub.com
KArnold
Active Member
Practically every flight I'm on the FA says "I'd like to welcome you to ....", and I wait. They would LIKE TO, but never actually do.
Think we need a new software release. V12.4?
Actually we NEED 12.4 to show up next week in a couple of (employee) cars on TeslaFi as proof it is starting and 12.3.7 released to entertain us in the slow 12.4.x rollout meantime.
I've seen this behavior several times. It's not perfect but compared to V11 basically infinitely better. I don't think V11 had this programmed. It appears to be emergent behavior from training. Another variant of this is not entering an intersection unless there is going to room to clear it. A few times I thought it was going to be wrong, entering the too early, but it either got lucky or it picked up on cues that I missed.
Thanks! I was first puzzled by those levels until I understood that they are not autonomous levels but accidents severityYes.
View attachment 1046461
Autonomous Car Progress
I think supervised FSD is a terrible idea, and is going to fall on its face so I agree with you on that one. I don't doubt that some version of FSD (or a competitors version) will be safer than the average human driver, but I don't think that's a good comparison. The entire comparison to...
But I could find the L4 number in that source
EVNow
Well-Known Member
EVNow
Well-Known Member
You are asking me to do your home work
Miles between incidents for regular cars (according to Cruise’s modified version of Virginia Tech study, which removes crashes on highways and rural roads):
Miles between Level 1 incidents: 282,485
Miles between Level 1 and 2 incidents: 99,404
Miles between Level 1, 2 and 3 incidents: 20,661
Miles between Level 1, 2, 3 and 4 incidents: 9,21(8)
Similar threads
