I agree with that, and that would be my concern as well. I would assume though that if they realized they needed the extra cameras, they would have added them in the S/X refresh.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
What if FSD doesn't materialize?
- Thread starter Goose66
- Start date
Tesla blaming the government is likely their best "out" with current hardware. I'm sure their developers are ready for the next gen computer. Ideally current FSD owners will get that computer. Not sure how Tesla would handle the situation of adding more sensors, including cameras.
I don't expect my Model Y to ever do level 3. But I'm quite happy with Autopilot.
The sun glare issue alone, which @Tam mentions above, may mean that even level 3 isn't possible
FloridaJohn
Member
Level 3 is entirely possible even with the sun glare issue. In a level 3 system, the car will just alert the driver that it needs to take over, and probably won't allow automated driving until the condition has alleviated.
Level 3 systems are automated until such a time it cannot handle the current conditions. The human driver is the backup to the automation.
I do agree, however, that level 4 or 5 may not be possible with a sun glare issue. Although, a fully automated car (without a backup driver) always has the option to pull over until it is able to continue on it's own. If the programming notices reduced visabilty due to the sun, it can just pull over for a couple minutes until the sun has moved position and then continue on its way.
Tam
Well-Known Member
What's missing above for L3 is the Safe "Handover/handoff" period/phase. For Tesla, that period is an unsafe zero seconds. It would just suddenly alarm disable its automation at any time including when its cameras are blinded by the sun glare. Another example of Tesla's unsafe zero-second handover period is phantom brakes. Sudden phantom brakes without warning do not technically shut down the system but it requires the driver to take over and press the accelerator.
L3 must give humans adequate time to take over. For Audi, it's minimum 10 seconds and the system starts to progress to an orderly and safe emergency procedure after no human response in 20 seconds. That also means no more phantom brake allowed. "Handover" period/phase can be as long as 45 seconds to warn humans to take over soon.
Many companies do not want to sell L3 because they can't figure how to implement a Safe "Handover/handoff" period/phase such as: How can they get rid of phantom brakes, sun glares?
And if they know how to eliminate phantom brakes, sun glares (Waymo, MobilEye...), they might as well progress to L4 instead of the troublesome L3 Safe "Handover/handoff" period/phase.
re: sun glare
I've had my Tesla for less than a year, but I was pleasantly surprised that AP worked well driving home (101 north) in the fall when the sun is right in my eyes. AP tracked the road perfectly, especially by the 85 exchange on-ramp where the asphalt is patchy and can be hard to follow.
I've had my Tesla for less than a year, but I was pleasantly surprised that AP worked well driving home (101 north) in the fall when the sun is right in my eyes. AP tracked the road perfectly, especially by the 85 exchange on-ramp where the asphalt is patchy and can be hard to follow.
FloridaJohn
Member
Yep. Totally agree.
I was really just responding to the comment that Level 3 wasn't possible because of things like sun glare. If Tesla can recognize an upcoming problem and if Tesla can do a smooth transition back to the driver, then Level 3 is possible.
I do agree, though, that if I was in charge of development I would completely bypass Level 3 for the reasons you mentioned. A development team would put a lot of effort into detecting potential problems, and verifying driver has again regained control. It might take just as much effort (or less) to tell the car how to handle the situation (i.e pull over) than it will to ensure a smooth hand-off.
I think if they were to do that it would only be in anticipation a future HW4(?) computer upgrade. My current understanding of the FSD hardware is that there are no extra camera-feed connections. So unless they're willing to consider expanding the input suite in an upcoming HW generation, it seems the most they can do is tweak the locations/angles/FOVs of the 8 cameras.
Well, speculating further, maybe they could multiplex an additional pair of camera bitstreams into one pair of the present setup. If the present HW3 is flexible enough to handle a bit of demultiplexing (perhaps alternating frame grabs between cameras, effectively halving each camera's FPS rate which is fine for this task) then the number of digital video connectors on the board need not be the ultimate bottleneck.
I'm not saying that Tesla engineers would consider additional cameras, or agree at all this could be helpful, but from here on the outside it sure seems like it could be. It would be interesting to know just what back doors (or roughed-in doorframes behind the wallboard) are available for such expansion in the platform architecture.
I've gotten it in the forward quarter before but I think only twice in the over three years of ownership. The cameras are able to compensate reasonably well but there can come a point where it gives up. Presumably it's like "ok I can't lower the ISO any more and I can't shorten the shutter speed any further either, let the meatbag take it from here"
No Tesla quite yet,,l but I'm quite interested in this because I drive due west every day around sunset.
Around 10 minutes before sunset, the sun can be almost directly coincident with the stoplights. It's so bad that I find myself planning my departure time to try to avoid this. But road glare and rear windows with reflective tint treatments can also be hard to handle, and I also hate driving in deepening twilight which is more challenging than regular nighttime. I hope the Tesla cameras can shrug off these challenges easier than I can, but it sounds like a problem.
Tam
Well-Known Member
Robotaxi from the movie Minority Report is fiction so it's nice to hear Tesla in 2016 saying that it could do it with all hardware there with the capability.
It's easy to say something that didn't exist then, but as Tesla started to work on the issue, I doubt that the hardware would stay the same throughout.
If they replaced the front side cameras (Rearward Looking Side Camera) with an anamorphic 180° lens, it might be an improvement. If there is such a video lens. Even 120° would be good. Anything to use that position as a forward-looking camera. Up the resolution on that camera to allow for the wider field of view.
Last edited:
Elon thinks we are close. Good enough for me. The problem can 100 percent be solved with vision. That’s how humans drive. The question is how long will it take for humans to achieve pure vision fsd.
pure vision fsd might not come first. The only question is how relevant is what comes out 2nd.
Personally I wouldn’t worry about Tesla achieving fsd until someone can do it on more that 50 percent of the roads world wide.
Others have replied as well, but I think Elon has said FSD is available by the end of 2021. FSD in this context should not be read as L5 FSD IMHO. It should be read as "the L2 driver assist system known as FSD".
Exactly -- FSD is just the name of the product, not the function, as with "dollar" stores. It's fine by the 95% plus of
us who never intend our Tesla to be used as a robotaxi, especially in a blinding snowstorm.
As for the marketing name, it reminds me of the lyrics of an old song by the Bonzo Dog Doo Dah Band
called "Shirt". In the song, someone asks how long a shirt would it take for a shirt to be cleaned
at a place called "59-minute Cleaners", viz. The Bonzo Dog Doo Dah Band – Shirt
Back in 2018 the Tesla website clearly promised a "door-to-door" full self-driving experience. Sorry, not going to cut it.
It's also entirely feasible to create a wide field-of-view using an array of two or three cameras and local stitching processor to create the one super-wide feed. This is not expensive these days and may be more effective than a fish-eye or anamorphic lens, particularly in rejecting glare and heavily directional illumination from sun, bright headlights and those intense night-construction worklamps.
So yes, this could be a solution that fits well into the current platform architecture, assuming there is some built-in flexibility with the exact configuration of the camera video format.
I think the fender cameras are mounted somewhat lower and Is less forward than optimal for the cross-traffic view, but maybe an improvement over the B-pillar. My choice would be high on the front corner or in the A-pillar.
Or have the car launch a mini-drone for a quick recce as needed.
But, how does the Robot feel about all this?
You're telling the Robot it has to be 10X better than a human, and yet you won't give it any additional sensors. You won't even give it sensor redundancy.
That's not fair.
You won't even give it down facing cameras that humans get in the cars they buy.
To me the entire point of FSD was an attempt to throw something together that could do some level of autonomous driving beyond that of what we had in 2016. They didn't really know what was needed as it had never been done before. I think they knew before they started that changes would have to happen. But, they won't make those changes until limitations are reached with the existing hardware.
The only mistake they made was the promise.
10x is easier than you think because the robot isn’t texting, putting sauce on his taco, or day dreaming about a milf.
Sure, but as has been discussed in other threads, that's going to be a matter for the courts at this point. There is basically zero chance we're going to get a L5 system released to the general public this year. I'm not saying that you shouldn't be angry that you're not getting a L5 system, but let's be honest, you aren't getting one this year and possibly ever.
Similar threads
