I’m on fsd beta too just like you. Have you noticed that it brakes a bit too late? It’s because it needs higher resolution cameras that can see further
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
-
Tesla's Supercharger Team was recently laid off. We discuss what this means for the company on today's TMC Podcast streaming live at 1PM PDT. You can watch on X or on YouTube where you can participate in the live chat.
There will be NO HW4 upgrade for HW3 owners
- Thread starter macmee
- Start date
M109Rider
Active Member
Again with the assumptions stated as facts. This is your thing I guess.
The media does that very same thing on a daily basis. Don’t be like the media.
The cameras see vehicles pretty far ahead. I know this as fact, because it shows them on the screen way down the road.
Is it possible that in an instance where the car might brake late, it’s not because it can’t see the vehicles, but for another reason ?
So will people with HW3 vehicles never see their robotaxi printing them $30k/year?
Will Tesla "value" HW3 FSD vehicles less on trade-ins? (Are they valuing FSD at all on trade-ins? Remember that conundrum from a year or two ago?)
They pretty clearly planted the flag in the ground that from 2016 on, all Tesla vehicles had the necessary hardware to become robotaxis. So will Tesla - assuming they ever take legal responsibility - have to underwrite HW3 vehicles differently than HW4 vehicles and differently than HW5 vehicles?
Of course, the main question is can a HW3 only vehicle reach 1,000% safer than a human while driving itself. If not, I smell a class-action lawsuit incoming.
Will Tesla "value" HW3 FSD vehicles less on trade-ins? (Are they valuing FSD at all on trade-ins? Remember that conundrum from a year or two ago?)
They pretty clearly planted the flag in the ground that from 2016 on, all Tesla vehicles had the necessary hardware to become robotaxis. So will Tesla - assuming they ever take legal responsibility - have to underwrite HW3 vehicles differently than HW4 vehicles and differently than HW5 vehicles?
Of course, the main question is can a HW3 only vehicle reach 1,000% safer than a human while driving itself. If not, I smell a class-action lawsuit incoming.
Ruffles
Member
Are you sure that is the reason? Sometimes it brakes later than I would but most times it is fine. Here again, I'm not convinced the input (i.e. cameras) are the problem and it's not a computational issue.
For people who thing the camera location/resolution is insufficient - do you really think no one at Tesla has suggested higher quality cameras or putting the cameras in the bumpers before now? I'm sure there was tons of research put into where the best placement would be and how much resolution they would need to be able to see at the required distances. Those are basic questions. Tesla has been doing this for years now and if there was some fundamental issues such as the cameras couldn't see, it would have been changed long ago before they started building up their library of training data.
The car brakes late because of latency in the processing pipeline, not because it cannot see the car ahead. Watch as a lead car, clearly identified, slows down to turn into a parking lot. Your car will continue too apply braking AFTER the lead car has completely cleared the lane. Your car sees the lead car just fine, but the latency means that your car does not know that the lead car is gone until after the latency time. I estimate this at about 0.5 seconds.
M109Rider
Active Member
Is it programming, or latency. ?
I’ve seen improvements that could have been seen as latency issues, but are now fixed. If those were latency, they wouldn’t be fixable.
Those would seem to be a programming fix.
I’m not sure either way. I’m not convinced it’s a latency problem though.
Some could be, who knows.
If you look at even the most recent FSDb release notes, you will see three references to improving latency. Earlier releases usually have references to improved latency as well. If latency is not an issue, why spend the effort improving it?
Clearly, there are issues other than latency, but it's a big factor in driving smoothness. With high latency, the ambiguity in position and velocity of dynamic objects grows quite large, especially at speed and the car must account for the worst case.
Another factor to consider is that the system is processing approx 30fps, which is 33ms per frame. How many frames does it take to determine what an object is doing? Is it static, is it in motion? If it's in motion, what is its trajectory and speed? If there is 0.5 second lag as suggested, that's 15 frames of data. There are multiple NNs and transformers in play. It might take 200ms through one NN to predict what's happening around the car (the occupancy network) and another 150ms for another NN to decide how the car should react to the situation presented (the planner), and another 150ms for the control systems to process and adjust the physical controls.for the car (accel, brake, turn, etc). The end result is that it seems to take 0.5 seconds to react to something.
The tweaking that can happen is using less frames to make a determination on occupancy, and lowering latency through the planner and control NNs. I still think they are going to need at least 5 to 10 frames for accurate positioning and relative motion determination. That's 165ms minimum (5 frames) without even handing off that data to other NNs for planning and control.
The tweaking that can happen is using less frames to make a determination on occupancy, and lowering latency through the planner and control NNs. I still think they are going to need at least 5 to 10 frames for accurate positioning and relative motion determination. That's 165ms minimum (5 frames) without even handing off that data to other NNs for planning and control.
I think the public perception is that all vehicles would come with the necessary hardware to become robotaxis, but this is an example of what was officially said
Source: All Tesla Cars Being Produced Now Have Full Self-Driving Hardware | Tesla
Full-Self Driving Capability might not mean Robotaxis. A safety level substantially greater than that of a human driver does not mean Level 4-5.
Nothing here references anything that relates to achieving generalized Robotaxis in what is required to achieve generalized Robotaxis.
Daniel in SD
(supervised)
How the heck would it "lower the financial cost of transportation for those who own a car and provide low-cost on-demand mobility for those who do not." if it's not a robotaxi?
There's also the fact that they explicitly said it was L5 capable hardware.
stopcrazypp
Well-Known Member
Use it for Uber? I believe there are people out there already doing just that using FSD Beta (which is squarely L2). Doesn't have to be a robotaxi (L4/L5) necessarily.
An investor conference call that probably a vast majority of buyers never heard isn't likely going to be successfully argued as a contractual agreement to deliver an L5 car for buyers (even a false advertising suit might be tough as the target audience was not car buyers). The SEC however might investigate it as securities fraud (as per recent news).
Above is just looking from the extreme end, assuming the worse on how Tesla may respond.
Last edited:
A couple months ago I was checking out the Tesla forecast from Cathie Wood's ARK and noticed this little gem
Source: https://ark-invest.com/articles/analyst-research/arks-tesla-model/
Maybe they expect Tesla owners to sit there and monitor L2 FSD while providing very low-cost on-demand mobility and fees to the company.
I'm not able to watch the full video right now but is there a specific timestamp in there mentioning L5?
I think there will be avenues like this through which to pursue action but it will likely take years and might not be a concern for a company now bringing in $80billion+ annual revenue.
ARK has long suggested Tesla trial a "human driven ride hail" with their vehicles first before going to robotaxis. Tesla has never suggested that would be the path.
AlanSubie4Life
Efficiency Obsessed Member
Well at least there’s still tons of upside even with zero robotaxi revenue! Probably good for them to focus on the car business anyway.
True dat! Tesla says 10ms but that has to be a best case theoretical scenario.
Last edited:
Yep, Elon keeps mentioning the robotaxis nonsense at the most recent earnings calls as well as lifted the FSDb's price ($15k) for HW3 equipped vehicles with that rationale. Still probably not enough to help compensate anyone in a court of law though.
Daniel in SD
(supervised)
@gearchruncher sued and won.
How to sue Tesla over historical claims
I sued Tesla in small claims and won. Twice! You can too. I had two claims: I have a car with AP2.5. Tesla advertised this car as having “all hardware needed for full self driving capability.” When I went to subscribe to “Full Self Driving Capability,” Tesla required a $1,000 hardware upgrade...
teslamotorsclub.com
In my view all of this is just a really unfortunate blemish on the car business that has otherwise done a lot of great things. I know people speculate about EAP/FSD helping the company survive during the most difficult periods and ok, maybe the ends justify the means but man oh man.
Tesla winning any kind of lawsuit depends heavily on their development behavior.
The stated claim is that the car has all HW needed for FSD CAPABILITY at a safety level significantly higher than a human.
If Tesla keeps developing for HW3, keeps doing software updates, and keeps feature sets the same as HW4, then there isn't much to argue here, because it will all be about the reliability. As you'll notice, Elon is being very careful to claim HW3 will still do FSD and will be safer than a person, HW4 will just be that much safter.
The instant Tesla makes HW4 more capable than HW3 in any way besides some baseline safety statistic, especially in any kind of area that could be argued to be fundamental to actual full self driving wihtout a person even in the car, they have an issue. I know I'll be watching this closely
The stated claim is that the car has all HW needed for FSD CAPABILITY at a safety level significantly higher than a human.
If Tesla keeps developing for HW3, keeps doing software updates, and keeps feature sets the same as HW4, then there isn't much to argue here, because it will all be about the reliability. As you'll notice, Elon is being very careful to claim HW3 will still do FSD and will be safer than a person, HW4 will just be that much safter.
The instant Tesla makes HW4 more capable than HW3 in any way besides some baseline safety statistic, especially in any kind of area that could be argued to be fundamental to actual full self driving wihtout a person even in the car, they have an issue. I know I'll be watching this closely
Similar threads
- Question
