EVNow
Well-Known Member
I don't think so. Its just the controls that are different. I don't think CT NN will be different from Model 3 NN.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: FSD Beta tweets
- Thread starter Terminator857
- Start date
DanCar
Active Member
If the front cameras are significantly higher than other vehicles then that might necessitate retraining.
EVNow
Well-Known Member
If this is really the case it would be terrible - because it will take a looooong time to get to billions of miles for just CT.
So, I think they probably just need to do some pre-processing of the camera feed - if really needed. I'm sure they are not waiting for a few thousand miles of training video they would be getting now per day.
I believe they can do an emulation mode, along the lines of Elon's comment that HW4 FSD is currently piggybacking on HW3 via camera* (and presumably procesor) emulation. They believe HW4 will eventually outperform HW3 FSD, but we won't see that until they apply the engineering resources to take advantage of the more powerful in-car inference computer.
* I do believe that the HW4 camera feed, even if down-rezzed to HW3 video, has some performance advantages.
If they were to start off using emulation to get to Cybertruck FSD they might use the bumper camera initially as a perspective-morphed fill-in to approximate the otherwise blind area in front of the vehicle. In other words, not taking full advantage of the view right in front of the bumper, but just to make up for the issue you mentioned.
Another possibility in all this, is that Cybertruck FSD will not have much to do with current-vehicle FSD, but will be an offshoot or parallel development of the coming and as-yet unknown Robotaxi FSD, which we might call HW5 FSD. But there are a lot of possible paths and variants here, and we don't know enough about the next-gen hardware to refine the speculation.
Regarding the idea that customized retraining per platform would be "terrible": I understand your point but I think that's an unduly strong statement. Again there are lots of possibilities, perhaps with base training accomplished through an existing library of video, possibly post-processed to fit, and incremental training clips that don't need to represent billions of CT-specific miles to achieve the goal. It doesn't have to be all or nothing in any aspect. To put it in human education terms, everyone gets the survey course and then they break up into smaller classes for advanced concepts or remedial ongoing reinforcement learning, tailored to needs and capabilities.
Elon gave a similar timeline for HW4 and support came much sooner than anyone expected. The optimist says maybe the same will happen with the CT. The pessimist says not a chance. My guess is by the end of July.
EVNow
Well-Known Member
What exactly is a "safety intervention?" Are they counting all the times FSD was disabled by the user? On my own experience, the number of times I did this for reasons that had nothing to do with safety is MUCH higher than it did. And even then, it wasn't "safety" as much as navigation errors: I had one problematic freeway interchange where it said it was going to take the left ramp and even signaled to do so and then stopped and headed back on to the "main" line: I had to kick it off quickly and manually navigate correctly.
diplomat33
Average guy who loves autonomous vehicles
Safety interventions are disengagements when you believed there was a safety risk, like when you thought FSD might cause an accident. For example, it looks like FSD might hit another car or might run over an object on the road etc... Disengagements due to navigation, routing, poor lane choice etc are not safety interventions. So the 1 in 380 miles is only counting interventions for safety risks, like to avoid an accident.
diplomat33
Average guy who loves autonomous vehicles
I am VERY dubious of this. Do you really think that even a dent has been put into the VOICE data given over the month to get any kind of solid, reliable information?
I don't.
Also, a LOT of people were using a new (to them) technology and most probably didn't understand it. Sure, you and me and most of the people on this board get it, but 95% of the population of Tesla drivers don't. People are panicky and will report issues that aren't issues. And you've been on this board long enough to know that people LOVE to scream "SAFETY ISSUE" over something that isn't.
I have a coworker who has had her 3 for close to three years and has yet to use basic AP as she is afraid of it and doesn't trust it. I've even taken her out twice to show her that it works great and is very safe. But because she has major trust issues, she won't use it. I'm sure she's not alone in this world.
diplomat33
Average guy who loves autonomous vehicles
I am not claiming the stat is 100% accurate. But since Tesla does not release official disengagement data, teslafsdtracker is the best we got. And it at least gives us a ball park.
And I agree that people may be over reporting issues as safety disengagements when they are not. So what do you think the actual safety intervention really is? 1 per 500 miles? 1 per 2000 miles? Are you suggesting that FSD (Supervised) actually is close to 1 safety intervention per ~10,000 miles needed to remove supervision? I doubt it.
The stats are helpful but everyone uses disengagements differently so it's hard to extrapolate what they mean. I think of critical safety engagements as a situation where I thought there was a good chance of an accident so I had to take over. I haven't had a critical safe engagement with V12 but had many with V10/V11. I'd be happy with L3.
Fully agreed. Add to that unless it’s an actual safety or error I like most report nothing. Half if not more of my disengagements are a desire to interject a “Plaid” thrill for fun or change direction from navigation. Like #aronth5 above I can’t remember my last danger disengagement.
At a guess, the voice feedback was more useful when drivers were prone to disengage before anything actually happened. A kind of anticipatory disengagement. "I don't know what it's doing, and I don't want to find out. Disengaged." Telemetry wasn't going to tell Tesla what was going on. With V12, it's far less useful for that, but can still give Tesla a glimpse of how people perceive the system during disengagements. Imagine if they filtered based on vulgar language and found that people get really upset for particular scenarios. That might spur them to prioritize work in those areas.
Perception is a big deal. The danger of the most recent free month is that since it wasn’t perfect, the majority of new users will now never use it in the future at any price because the “trauma” is now locked in.
For myself, I almost never use it because it is so stressful constantly disengaging it, worrying that you could wreck at any moment. I disengage for any reason on average every 5 miles or so, so there really is no enjoyment for me. If I could, I’d refund at this point. Otherwise, I’m consistently inching toward never again at any price territory.
For myself, I almost never use it because it is so stressful constantly disengaging it, worrying that you could wreck at any moment. I disengage for any reason on average every 5 miles or so, so there really is no enjoyment for me. If I could, I’d refund at this point. Otherwise, I’m consistently inching toward never again at any price territory.
Many people who I have talked to who tried the free month were in general pretty positive. Common feedback was they would very likely subscribe after FSD was upgraded a couple of times. Many want to wait for the 8/8 unveiling before deciding. Nobody mentioned anything about "trauma". Just the opposite they were surprised how much FSD had improved since most had previously experienced FSD with friends/family.
EVNow
Well-Known Member
There are always going to be early adopters, mainstream users and laggards.
FSD is still definitely in the bleeding edge / early adopters phase. You are not going to be adopting it until the disengagement rate goes down a lot.
ps : I'd like to see how a driving teacher would use FSD. They are the ones who need to do this intervention / disengagement with humans.
willow_hiller
Well-Known Member
Improvements in responding to hand signals coming in 12.4 and 12.5:
EVNow
Well-Known Member
Why not school zones ? I don’t know why nobody among his YT supporters asks him that.
Similar threads
