Lets put this to the test and see how many people STILL fall for Elon's tall tales. Like I accurately predicted Karpathy would NOT return, putting it in writing here: FSD will NOT be solved with just vision.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Another of my predictive polls: FSD will or will not be achieved with "vision only". VOTE NOW!
- Thread starter 2101Guy
- Start date
Your poll does not have a timeline and is not specific enough.
FSD can be achieved with vision only.
Safe deployable L4 - L5 FSD will not be achieved in the next 3- 5 years with vision only and by vision only I mean the low resolution 1.2mp sensors that were sold on Teslas.
FSD can be achieved with vision only.
Safe deployable L4 - L5 FSD will not be achieved in the next 3- 5 years with vision only and by vision only I mean the low resolution 1.2mp sensors that were sold on Teslas.
diplomat33
Average guy who loves autonomous vehicles
With those limitations, Ii should be obvious that "vision-only" is not the smart approach to achieve safe and reliable driverless everywhere. And Tesla's hardware of only 8 1.2 MP cameras and a low end computer, will NOT achieve safe and reliable driverless everywhere. So yeah, Elon was full of BS when he claimed AP2 was "L5 capable".
Now, if the goal is a driver assist, vision-only is perfectly fine since you have a human driver as back-up if the system makes a mistake.
I think Tesla will get around this by keeping driver supervision for FSD. So Tesla will put out a driver assist that does "door to door" and requires driver supervision but they call it "Full Self-Driving" so they can technically say that they achieved FSD with vision-only. But it won't be what most people consider true FSD.
Source:
What Is Lidar? 10 Things You Need to Know About Laser Sensor Technology
Nearly all major developers of safe self-driving technology use it. Here are the top 10 things you need to know about lidar.
groundtruthautonomy.com
Last edited:
Daniel in SD
(supervised)
This is a dumb poll. FSD (robotaxi) will be solved with vision only. It's just not going to be solved with HW3. Now by the time it can be solved with vision only (50 years? who knows) we may have entirely new sensor technologies that we can't even imagine now that are even cheaper than cameras. So maybe no one will actually do it for any reason other than a science experiment. 
Vision only or lidar or radar all require a current knowledge of speed limits to be functional. Unposted speed limits in rural areas are far from accurate and Tesla offers no mechanism to correct it. After a recent long conversation with Tesla Service, they claim that the speed data comes from municipalities, and if incorrect, the burden is on us to get it corrected. They offer no help with this. They can’t even tell us who to contact. Without accurate speed limits, none of this works, regardless of the sensors. Bad maps = no ADAS.
jebinc
Well-Known Member
Not many. FSD is a scam that stole $12k from many. EAP is more accurate. Still over-priced, but does what is promised. Not selling a false promise. Malicious or otherwise (FSD false promises).
This translates to "I believe Elon is lying (again). Just for other reasons".This is a dumb poll. FSD (robotaxi) will be solved with vision only. It's just not going to be solved with HW3. Now by the time it can be solved with vision only (50 years? who knows) we may have entirely new sensor technologies that we can't even imagine now that are even cheaper than cameras. So maybe no one will actually do it for any reason other than a science experiment.
Which essentially qualifies as a "no" vote. Feel free to voice your vote!
Daniel in SD
(supervised)
Nah. Because I think that by the time AI technology is capable of driving better than the average human everywhere then it will also be able to do it with vision only.
bradtem
Robocar consultant
The problem with this poll is "will be solved." Most people, even the ardent LIDAR fans, will say that some day it should be possible to solve it with vision only. But that's not the problem before anybody. The problem is the quickest way to solving it in a commercial product. And almost everybody but Tesla believes that vision is not the quickest path.
Lidar/radar doesn't read speed limit signs like a human can, but a camera should be able to easily. In California, an unposted speed limit means use the "basic speed law" which currently is 55mph or safe for conditions (fog/rain etc). Lacking any other valid source, the driver would have to input it.
Last edited:
AlanSubie4Life
Efficiency Obsessed Member
I agree this is a dumb poll. I'm a proponent of good polls which determine what people actually believe.
FSD solved to MOST, equates to what the CEO said it means: NYC to LA with NO human input/intervention.
But if thats too hard to figure out? Then dont participate or vote in the poll. Simple.
Life really isnt all this hard.
But if thats too hard to figure out? Then dont participate or vote in the poll. Simple.
Life really isnt all this hard.
AlanSubie4Life
Efficiency Obsessed Member
Me neither
Crappy poll is crappy
Edit: I don't have anything against the OP as we tend to agree on a lot of things FSD related, but common this poll is crap.
My question is why vision only? If I can add in 2 side radars and a front 3D radar for $250 per car, then that can only make it even better than vision only. That’s assuming that you don’t believe Elon and think that the AI can be developed to fuse the different sensor data (like it already does in many other implementations).
1. Because it was cheaper and not economical at that time to add expensive sensors to the car that was already costly to produce and difficult to make.
2. They have sold cars with promises of L5 autonomy using vision only because it is a very compelling selling point as most on this forum can attest to
3. A big part of Tesla valuation is based on that as well.
Now they are just trying to make it work. The CEO underestimated how monumental a task they are trying to solve and overestimated their capability to make progress and deliver on the promise and product that was sold.
In theory, fusing multiple sensors is better but if you have inadequate sensor coverage, a sensor package that was not designed for ADS and you generally don't know how to fuse the sensors properly then it just won't work out. Sensor fusion is currently the only way to deploy a safe L4 ADS because vision alone has certain limitations and risks that no car maker is willing to take. Especially at L3 - L5 which means you take responsibility if anything goes wrong.
jebinc
Well-Known Member
Similar threads
