Coast to coast drive happening this year for all FSD Teslas!

[diplomat33]

Were you at the actual event? What kind of actual proof are you referring to?

I am referring to all of it, including the testimonials from test drives which said that Tesla had good FSD able to handle local roads and intersections.

 

[CK_Stuggi]

Ok, thanks for the clarification. I'm hopeful this time is different, but I wouldn't quite go as far as calling it proof.
I'm just sceptical because it was a big event, with many explanatory words and a YouTube video.
Since then 2 weeks have passed and not a single new feature has been rolled out (rather, advanced summon was delayed and bugs lowering performance were found).
Let's hope the autopilot team shakeup electrek reports on just now will not delay things.
Tesla, please - prove my scepticism wrong!

 

[diplomat33]

Hopefully, Tesla not just does the coast to coast FSD but also releases the FSD software like Elon said so that we can duplicate the demo ourselves. Maybe the naysayers will finally shut up about "where's the coast to coast demo?"

 

[buttershrimp]

You had to say it...

Adult Movie Filmed In Tesla Autopilot And We Are Wondering If It's Safe

Well, it can’t be Mad Max without the the semi truck full of ladies.... welcome to Valhalla! Someone please post this on their thread!!

 

[pilotSteve]

Then it's called AP and not FSD. It's not self driving, much less FULL, if the owner is responsible for supervising it. Then it's just ADAS still.

Yeah, I mean WTH did Elon have to over-sell it with "FULL" from the very first announcement? "Self Driving" wasn't enough????

 

[NYCR160]

Coast to cost will never happen unless they have new software for all of us. Driving in moderate rain kicks it down a notch from NOA to TACC. In snow or heavy rain TACC turns off. Whoever is doing the FSD software needs to turn off their simulators and start driving their software in real weather.

 

[diplomat33]

Coast to cost will never happen unless they have new software for all of us. Driving in moderate rain kicks it down a notch from NOA to TACC. In snow or heavy rain TACC turns off. Whoever is doing the FSD software needs to turn off their simulators and start driving their software in real weather.

They do have better software as evidenced by what we saw during Autonomy Investor Day and the FSD demo and test drives. I am pretty sure AP3 won't get kicked out under heavy rain like the current NOA does. AP3 has better neural nets that will be able to process what the cameras see.

 

[Watts_Up]

They do have better software as evidenced by what we saw during Autonomy Investor Day and the FSD demo and test drives. I am pretty sure AP3 won't get kicked out under heavy rain like the current NOA does. AP3 has better neural nets that will be able to process what the cameras see.

Beside bad weather limitations, sunny days can also be a issue for detecting red light, stop signs...,
like being between Scylla and Charybdis or a rock and a hard place when the sun also ariseth, and the sun goeth down...

 

[diplomat33]

Beside bad weather limitations, sunny days can also be a issue for detecting red light, stop signs...,
like being between Scylla and Charybdis or a rock and a hard place when the sun is on the horizon.

I doubt that will be a problem for AP3.

 

[iTech]

Beside bad weather limitations, sunny days can also be a issue for detecting red light, stop signs...,
like being between Scylla and Charybdis or a rock and a hard place when the sun also ariseth, and the sun goeth down...

The sun isn't actually an issue for the forward facing cameras. @verygreen has a picture/video about it.

 

[acoste]

The sun isn't actually an issue for the forward facing cameras. @verygreen has a picture/video about it.

Did you see this video? Sun behind the car makes things difficult. But given that others have solved this issue, I guess Tesla will figure this out as well.

"nice touch - if it is unsure of current light status - it would approach the intersection carfully so it can stop gently should it actually be a red light - see at 6:26"

"So now the problems. it's actually not very robust at telling red from green in some conditions, see at timestamp 4:05 how it flipflops. and then nearly runs a red light at 5:12 and I need to intervene."

 

[NYCR160]

I will give them the benefit of the doubt for making it work in heavy rain. I am going to remain highly skeptical about their ability to keep it working in the snow. Given who quickly it gave out last winter in any snow event, I would consider it only a fair weather FSD system.

 

[diplomat33]

I will give them the benefit of the doubt for making it work in heavy rain. I am going to remain highly skeptical about their ability to keep it working in the snow. Given who quickly it gave out last winter in any snow event, I would consider it only a fair weather FSD system.

Interestingly, I think Elon or Karpathy actually mentioned during the Autonomy Investor Day presentation that they are working on getting it working in snow and it works pretty well. I guess the vision neural net can detect the tracks in the snow from other cars or maybe detect the difference between packed snow on the roads versus undisturbed snow on the side of the road but they said the vision actually can predict correctly where the road is even covered under snow.

 

[NYCR160]

If and when FSD works on snow cover roads and in a snow storm it will then making prepaying for FSD worth it. Right now I am frustrated with its inability to work well in the rain which happens every other day now.

 

[Wooloomooloo]

They do have better software as evidenced by what we saw during Autonomy Investor Day and the FSD demo and test drives. I am pretty sure AP3 won't get kicked out under heavy rain like the current NOA does. AP3 has better neural nets that will be able to process what the cameras see.

AP3 doesn't have "better neural nets" it simply has more compute ability. Solving edge cases such as weather or other anomalous conditions isn't (at least initially) a compute problem, it's a data curation and learning problem as well as a software problem. If AP2.5 could do sunny day FSD and AP3.0 was required for rainy day FSD, we'd have seen sunny day FSD alerady.

People are putting way too much faith in Tesla being able to "compute" their way out of it. There are already far more powerful computers out there (in data centers) that still cannot do autonomy. If compute were the only issue, we'd just have really slow autonomy - like you said in your first sentence, it's about the software.

 

[willow_hiller]

AP3 doesn't have "better neural nets" it simply has more compute ability. Solving edge cases such as weather or other anomalous conditions isn't (at least initially) a compute problem, it's a data curation and learning problem as well as a software problem. If AP2.5 could do sunny day FSD and AP3.0 was required for rainy day FSD, we'd have seen sunny day FSD alerady.

People are putting way too much faith in Tesla being able to "compute" their way out of it. There are already far more powerful computers out there (in data centers) that still cannot do autonomy. If compute were the only issue, we'd just have really slow autonomy - like you said in your first sentence, it's about the software.

AP3 doesn't have better trained networks yet, but it will very soon. AP2.5 has been running on compressed footage, and more computing power will let them process uncompressed footage and presumably create deeper, richer neural nets.

 

[Wooloomooloo]

AP3 doesn't have better trained networks yet, but it will very soon. AP2.5 has been running on compressed footage, and more computing power will let them process uncompressed footage and presumably create deeper, richer neural nets.

Sure, I have no doubt the additional fidelity will allow the individual algos to discern information not possible today, your second point here is a great one and likely very true - and with that, there will be a lot of better filtering of noise caused by visual issues exacerbated by weather.

I think you'll find that people who have some background in this topic are less convinced than those who do not. Regardless, what's invisible to all of us is how well Tesla's classification and clustering is working in the back-end, on the software side. Compute will help with the latter, assuming the classification is solid in the first place, as well a regression or other reasoning and decision making algorithms. However, we don't know much about the classification aspect of Tesla's process upon which everything else hinges. Maybe it's really great!

 

[diplomat33]

AP3 doesn't have "better neural nets" it simply has more compute ability.

Back in 2018 I think it was, Karpathy said that Tesla already had, at that time, larger and more complex neural nets that AP2 could not handle. So AP3 can handle these larger NNs. So, yes, AP3 will have better neural nets.

 

[EVNow]

Hopefully, Tesla not just does the coast to coast FSD but also releases the FSD software like Elon said so that we can duplicate the demo ourselves. Maybe the naysayers will finally shut up about "where's the coast to coast demo?"

If FC is end of the year, I'd expect city NOA (basically what we saw in the demo) to be released Q1 next year.

 

[Wooloomooloo]

Back in 2018 I think it was, Karpathy said that Tesla already had, at that time, larger and more complex neural nets that AP2 could not handle. So AP3 can handle these larger NNs. So, yes, AP3 will have better neural nets.

Do you have a link to his statement, I'm curious about the context in which it was said.