Why FSD will never fully work on current cars

[Mod3forMe]

This post makes no sense at all, I mean no sense.

Do you drive your car with snow on the windshield... No... Do you drive without defrosting a foggy window or ice... No... You fix the problem so you can see.




Why is this different...........

I can clean my windshield while I am driving. I can not clean the cars camera and sensors while driving you - have to stop and get out to do this. E

And you seem to miss the whole point of the post. Elon/Tesla claims the model 3 has all the hardware needed to drive without anyone in the car. How can this work, when getting splashed by muddy water in the side of the car turns off auto pilot?

 

[GateFather]

I think FSD is the price of entry for maybe not true level 4 or 5 FSD (at least not this year despite Elon's confidence), but for step improvements until we get to that point. As new and smarter features become available and Tesla needs to train the ML algorithms, the FSD folks will be the ones training those models and contributing to the Tesla hive mind. Even if the Software we have on our car right now is WAY less capable than some much better version Tesla has on their test cars and they plan to shock the world with what they really have ready, the regulators in most countries are not going to approve level 4 or 5 FSD anytime soon. Maybe they can make these claims of feature complete FSD by end of 2019 because they know regulators will turn it down and they can use that as a "it isn't us, it's them, we held up our end of the bargain"

 

[JeffK]

I've been splashed by snow, water. and mud and I've never had autopilot disengage like you're suggesting. I have had NoA go to regular AP in poor weather conditions, but that's about it.

 

[Daniel in SD]

I think FSD is the price of entry for maybe not true level 4 or 5 FSD (at least not this year despite Elon's confidence), but for step improvements until we get to that point. As new and smarter features become available and Tesla needs to train the ML algorithms, the FSD folks will be the ones training those models and contributing to the Tesla hive mind. Even if the Software we have on our car right now is WAY less capable than some much better version Tesla has on their test cars and they plan to shock the world with what they really have ready, the regulators in most countries are not going to approve level 4 or 5 FSD anytime soon. Maybe they can make these claims of feature complete FSD by end of 2019 because they know regulators will turn it down and they can use that as a "it isn't us, it's them, we held up our end of the bargain"

So they’re going to not release it in the US because it won’t be legal in some other country?
Elon Musk himself said they’re only 98% done at the last earnings call. To me that sounds like they’re where the state of the art was over a decade ago. Obviously they’re doing it at way lower cost which is good.

 

[JeffK]

Elon Musk himself said they’re only 98% done at the last earnings call.

While people from Waymo usually describe themselves as 90% done with 90% to go.

 

[GateFather]

So they’re going to not release it in the US because it won’t be legal in some other country?
Elon Musk himself said they’re only 98% done at the last earnings call. To me that sounds like they’re where the state of the art was over a decade ago. Obviously they’re doing it at way lower cost which is good.

I didn't mention the US at all. In the countries where they sell, US included, I am making the point that I'd be shocked if regulators allow even level 4 autonomy by the end of 2020, let alone 2019. I was trying to make the point that they may feel comfortable in their claims from an accountability perspective because while they may or may not be ready in the timeframe they are advertising, they can always fall back to regulators lack of approval as the reason for delay instead of their own unreadiness.

 

[Daniel in SD]

While people from Waymo usually describe themselves as 90% done with 90% to go.

Tesla can almost do Musk’s commute while Waymo can go 11k miles between disengagements. I guess Musk may not have the super secret software build people keep talking about though.

 

[GateFather]

While people from Waymo usually describe themselves as 90% done with 90% to go.

Yep and I think as you near 100%, the 90% to go exponentially gets harder to cover. (i.e. 95% with 85% to go or 98% with 80% to go). Those last couple percentage points will be tough. Getting to 4 or 5 9's of accuracy (99.99% or 99.999%) will be even harder.

 

[Daniel in SD]

I didn't mention the US at all. In the countries where they sell, US included, I am making the point that I'd be shocked if regulators allow even level 4 autonomy by the end of 2020, let alone 2019. I was trying to make the point that they may feel comfortable in their claims from an accountability perspective because while they may or may not be ready in the timeframe they are advertising, they can always fall back to regulators lack of approval as the reason for delay instead of their own unreadiness.

Autonomous vehicles are legal in much of the US including California where half of Teslas are sold. Here is the form to register an autonomous vehicle in CA:
https://www.dmv.ca.gov/portal/wcm/c...89c5-b2bc7de3fd2c/ol321.pdf?MOD=AJPERES&CVID=

 

[JeffK]

I'd really like to see more demos of anything Tesla has. It's been two and a half years and we haven't really seen anything new. I wouldn't really consider NoA game-changing or anything. It's just an incremental improvement of AP.

 

[GateFather]

Autonomous vehicles are legal in much of the US including California where half of Teslas are sold. Here is the form to register an autonomous vehicle in CA:
https://www.dmv.ca.gov/portal/wcm/c...89c5-b2bc7de3fd2c/ol321.pdf?MOD=AJPERES&CVID=

True but if it’s not legal in all states then depending on where you live, which states you border, and what their laws are you may be in a less than ideal situation. I’m cautiously optimistic but I’m also a realist. I live in NJ which as far as I know has not passed any laws allowing autonomous driving. Bummer for me. I did purchase the FSD package for $2,000 though in the hopes that it’ll be available some day.

 

[Daniel in SD]

True but if it’s not legal in all states then depending on where you live, which states you border, and what their laws are you may be in a less than ideal situation. I’m cautiously optimistic but I’m also a realist. I live in NJ which as far as I know has not passed any laws allowing autonomous driving. Bummer for me. I did purchase the FSD package for $2,000 though in the hopes that it’ll be available some day.

I wouldn't worry too much about it. haha. Tesla has a large enough installed base that if they release Level 3-5 software it will quickly cover enough vehicle miles to prove its safety. I expect it would be very quickly approved in all 50 states if it proves far superior to human drivers as they claim.

 

[EV-lutioin]

Will be solved with a software update.

 

[aronth5]

I've been splashed by snow, water. and mud and I've never had autopilot disengage like you're suggesting. I have had NoA go to regular AP in poor weather conditions, but that's about it.

Well many owners, myself included have had EAP disengage with "weather" warnings on the screen. So if Tesla programmed a warning for the driver to take over due to bad weather what does that tell you!

 

[GateFather]

I don't think the current Model 3s are meant to eventually be the final version that will one day replace user driven cars in total. I think with the current sensors and cameras, and the upgraded computer, the Model 3 will be the first FSD autonomous vehicle mass produced and sold in high volume which will support FSD, if the conditions are right. So, you might not be able to use FSD all the time - to OPs original point.

I'd imagine that by the time vehicles are being produced which are true level 5 autonomous, they will have solutions figured out for sensor and camera fidelity regardless of outside road and weather conditions and will be able to operate fully independently without passengers all the time.

Taken even a step forward, I believe that sometime in the next 30 to 50 years owning a vehicle will be unnecessary and impractical. People will likely still own their own cars but they will flip to the minority. Transportation will be handled like Uber without drivers. There will be just as many cars on the road, but the majority of them will be autonomous and shared. They will be readily available to pick you up and take you to your destination for a very reasonable cost. Maybe in 100+ years driving your own vehicle will become illegal and owning your own car will be for collection and historical purposes. Just some fun food for thought.

 

[dnvr-ev]

I've been splashed by snow, water. and mud and I've never had autopilot disengage like you're suggesting. I have had NoA go to regular AP in poor weather conditions, but that's about it.

Autopilot doesn’t work in real snow. The sensors get covered and if you engage it, steering doesn’t work right.

 

[Mod3forMe]

I've been splashed by snow, water. and mud and I've never had autopilot disengage like you're suggesting. I have had NoA go to regular AP in poor weather conditions, but that's about it.

I had this occur about 10 times. Given I have only EAP for 30 days and AP for 5 days that a fair bit.

Frost is the main culprit but like I said Dew caused an issue, sleet caused an issue, snow of course caused an issue and one getting getting splashed caused an issue.

Getting splashed and needing the driver to take over was the most interesting one. If the car gets partly is blinded I have to wonder what the FSD would do if driving on its own as it would not be able to safely pull over.

(Please remember the premises of this post is Tesla/Elon claimed that current M3 have the hardware to drive around without a driver - I don’t believe this is possible.)

 

[aronth5]

I suspect that Elon would like to go back in time and change his statement now that there is a lot more EAP feedback in their neural network.

 

[Daniel in SD]

Getting splashed and needing the driver to take over was the most interesting one. If the car gets partly is blinded I have to wonder what the FSD would do if driving on its own as it would not be able to safely pull over.

Most of the time it could probably safely pull over. If not it would have to stop in the middle of the road. It wouldn't be the first time a Tesla has broken down
Of course, being in California, I've never had autopilot disengage due to obstructed sensors. It does seem likely that FSD will not work in all conditions.

 

[Mod3forMe]

Most of the time it could probably safely pull over. If not it would have to stop in the middle of the road. It wouldn't be the first time a Tesla has broken down
Of course, being in California, I've never had autopilot disengage due to obstructed sensors. It does seem likely that FSD will not work in all conditions.

I guess it depends on the what got blocked. I am just thinking if the car is effectively blind on one side pulling over could be a challenge for it.

Funny enough I had a large bird crap on my car at stop light. It missed the pillar camera but was close to it. Something silly like that could blind an autonomous car - making the whole regulating them thing a challenge. Humans can (mostly) deal with the odd an unexpected - computers, not so much.