Forums

FSD on City Streets

FSD on City Streets

I may have missed something, but tried searching online and through the official forums search (aka google). I thought v10 was supposed to add Full Self Driving (FSD) on City streets, or at least basic recognition of stop signs and stop lights.

I can still find reports from previous months about those in the Early Access Program (EAP, not to be confused with Enhanced Auto Pilot which is also called EAP). This was images showing it detected a stop light or navigating on autopilot from the persons "home" to a destination including both city and highway driving. Yet v10 is here and there is nothing in the notes. Using Navigate on Autopilot still warns you it is ending as you near an off-ramp (and still randomly speeds up as soon as it turns off right before going into the turn for the off-ramp).

Don't get me wrong, the new visuals are much improved (though I still haven't seen or at least noticed a pickup truck appearing as a pickup truck), the cars dance far less and it seems to detect cars further ahead and in oncoming lanes better (though seemingly only about 1 out of every 3 cars in on coming traffic about the same for 1 out of every 3 people near the car). Though even with these visualization improvements the car will still dramatically slow down randomly when driving next to another car on the highway. I am also being prompted more often to confirm I am holding the wheel by shaking it, though most of the time when I shake it is not enough to clear the message or it turns off autopilot because it was too strong (yet to consistently find what it is hoping for). This wouldn't be much of a problem except for I have had it happen on the highway several times in under two minutes, which actually takes my attention off the road and onto the screen. Well that and I am driving with both hands on the wheel. The system definitely seems to function better with only one hand on the wheel as it creates a weight differential and thus more resistance on the steering wheel (the irony that someone attaching a weight to their steering wheel is thought by the car to be someone paying more attention than someone who actually has both hands on the wheel is almost comical).

I am straying from my point though. The two features I was expecting from this biggest every update and justification for a FSD price increase were supposed to be Navigate on Autopilot on city streets (recognizing stop signs and lights, and making turns) and smart summon. We got smart summon but where is the rest. Hopefully coming in an update soon, but then hard to predict what will be in an update. I thought the first two numbers in the firmware was year and week number. Well we are on week 41 and gotten several updates to what I would assume is a week 32 update. With all of the data from everyone beta testing smart summon one would thing we might be at least on week 38 or 39 by now with all of the data crunched from people testing out this feature over a million times in a week.

Hopefully someone has more information about this than me.

NKYTA | 13. oktober 2019

Soon.

EVRider | 13. oktober 2019

I don’t think anyone from Tesla said the initial release of V10 would support FSD in the city or respond to stop signs and red lights. The V10 release notes in the car and those on the support page (https://www.tesla.com/support/software-version-10-0) don’t mention those features.

Darthamerica | 13. oktober 2019

I don't think Tesla is anywhere near ready to release FSD on city streets. Not that they wouldn't based on history. But it would be a big mistake to go too early in scenarios like city streets. Let's get Smart Summon working safely first...

Techy James | 15. oktober 2019

FSD in City Streets would require 2 items. 1) The FSD feature set to be completed which by Elon's Optimist Timeline would be end of year 2019. Meaning no release likely to early 2020. 2) Cars to have FSD Hardware 1.0 upgrade.
The V10 was never listed as FSD feature other than the Smart Summon. Oddly even though this was new FSD I believe car with the Old Enhanced Auto Pilot (EAP) feature also got this new feature. Making me wonder when Tesla will draw the line on the old EAP versus FSD.
As a person that paid for both EAP and FSD I just want to see a benefit for the extra 3K purchase I did before the change from EAP/FSD to AP/FSD. If it is City driving then fine, but if that is case, then Tesla's argument from raising price of FSD by a 1K is bit early as they still haven't delivered a FSD specific feature.

Darthamerica | 15. oktober 2019

I think the car will need new hardware for FSD. It will get close on the freeways, but the car just isn't aware enough to be safe on a city street without hands on the wheel and brakes. I think it is technically possible to use vision based systems. But it will take many many years of SW development to implement. I think we underestimate what our brains do when we drive. Just think of when you hear an emergency vehicle. You automatically start slowing or taking precautions. A Tesla wouldn't react to that and could go right through a green light and get T-boned.

jimglas | 15. oktober 2019

Darth, your car can process more information faster and more accurately than you ever will. Its already a safer driver than you are.

Darthamerica | 15. oktober 2019

No it is not. No it can't. We are a long time away from that.

jimglas | 15. oktober 2019

i dont know if you are willfully ignorant or just making stuff up

Darthamerica | 15. oktober 2019

No autonomous car could even pass a DMV test. They are not ready nor will they be for a long time. The human brain is orders of magnitude more powerful computationally. If you doubt that, Climbing to the backseat of your car while it is driving and go to sleep will change your mind! Oh wait, it won’t let you do that!

jimglas | 16. oktober 2019

"The human brain is orders of magnitude more powerful computationally"
utter nonsense

Darthamerica | 16. oktober 2019

You're naive if you think Tesla has a computer as capable of driving vs a human brain. You truly do not understand the magnitude of this challenge.

TeslaTap.com | 16. oktober 2019

I'm not aware of any human brain that can drive. Now a human with a brain can often drive (but not always). Now a computer can't drive either, but a computer, with validated software in a car designed for it can drive.

It's already here today in Teslas for the freeway (although it does require human attention). My best guess is FSD will arrive in less than 18 months. Google has it today and is authorized to drive in CA without a human driver. Clearly not 10-years away.

I understand some without any technical background doubt this, but I've been following this industry for the last 5 years, we are far closer than most realize. That doesn't mean FSD will handle every single case. There will be some geo fencing, and it will not handle conditions no human should drive in either (whiteouts, hurricanes, lava flows, etc.). It will handle 99.9% of daily driving.

Darthamerica | 16. oktober 2019

TeslaTap you're greatly underestimating, just like Tesla, the magnitude of the problem. Tesla has done this before with AP1 and it had consequences. We have nag because of it. Simply put, Tesla's aren't able to percieve their surrounds or the context of their surroundings will enough to realize true FSD. This is going to take years more of NN learning and SW development if the intent is to rely primarily on vision. The data is there but making sense of it isn't easy to implement in a rules based code. Especially when the inevitable mistakes will result in regulations that cut off your development path. We will get to something closer to FSD on freeways. But otherwise it's going to be a a long long time and there will be a "man in the loop". Even if the technical challenges are solved the regulations to allow it to reach the promised potential will lag by years. Again, Tesla will likely push what it calls FSD early and at a grossly immature state. But it won't end well.

nukequazar | 16. oktober 2019

@jimglas, I'm not sure what cars you're driving or how well your brain is functioning but Tesla OS does not come anywhere close to a human of below average intelligence. Sure, it has sensors but it is nowhere near synthesizing information to make decisions anywhere near on par with basic functional human intelligence.

Maybe most people haven't noticed but so far FSD is a scam. It's nowhere near "full self-driving" or autonomous driving. At this point it is simply advanced cruise control with lane keeping and lane changes.

As @Darth touches on above, I don't see a computer in a car being anywhere near capable of autonomous driving in our lifetimes. We may get close with the help of neural networks once a threshold of enough cars on the roads communicating with each other that all their sensors will work together to guide more and more advanced cruise control. True autonomous driving, however, I think would require a DOD-level super computer running realtime inside the car (of course not possible) with dozens or hundreds of 3D sensors with LIDAR all around the car at multiple levels.

nukequazar | 16. oktober 2019

All this said, it makes me feel really good about human intelligence. Even an uneducated teenager of below average intelligence is an infinitely more capable driver than Tesla FSD.

NKYTA | 17. oktober 2019

Let's summarize DA and nuke.

Led ins: I own a Tesla. I really like my Tesla.
Tesla has problem.
Porsche EV is better (by whatever standard of the day).
Climate science is bunk, because I said so.
I'm smarter than the engineers and scientists at Tesla.

Continue trolling and we'll keep calling you out when you keep stating that 19 is greater than 240.

Darthamerica | 17. oktober 2019

OK sooo this isn’t about climate science, so I we can table that for now unless you persist. Now point by point... 1. I’m not saying “Porsche is a better EV”, that’s subjective. I am saying Porsche’s EV is a better car for racing and has better performance. 2. I am professionally as qualified as Tesla “engineers and scientist”. I’m not discounting their experience with building cars, but I do have enough experience to see the issues they have or to understand why they build their cars the way they do. This is why I knew immediately that Taycan would outperform Model S on the track. 3, with regard to FSD, a rat is more intelligent and better at obstacle avoidance than Tesla’s cars. This isn’t a know on Tesla, its just an acknowledgement of the primitive state of the art with respect to self driving cars. As Nuke said, what we are calling FSD is just a really good implementation/integration of cruise control + lane keeping + GPS navigation. But it isn’t FSD. Emphasis on FULL self driving. There’s nothing “full” about it. At best Tesla is close to limited man in the loop “Conditional Self Driving”. The visual data is there. But processing and putting into context that data in a driving scenario is well beyond todays state of the art. Until the cars can become more aware of the environment rather than just in it, we can’t have FSD.

jimglas | 17. oktober 2019

wrong
wrong
and wrong

NKYTA | 17. oktober 2019

"I am professionally as qualified as Tesla “engineers and scientist”."

Well, if you say so. Sure.

Darthamerica | 17. oktober 2019

The trouble for you is that you wouldn't know. That's why you're buying into fantastical promises that aren't yet achievable. There's a reason why the more technically competent someone is, the less likely they are to accept Tesla's FSD claims at face value. I'm not saying it's impossible, just not likely anytime "soon".

sunsinstudios | 18. oktober 2019

isn't there a video on tesla's youtube channel where the car drives itself?

I remember when tesla first started 10ish years ago; the question then was can electric cars can be s3xy? Now we are talking about whether the car can drive itself on the streets as well as it can on the highway. Soon is sooner than "soon".

Madatgascar | 19. oktober 2019

Yes, there’s a video of a Tesla driving itself through light traffic in Palo Alto. The FSD system would probably really work that well for everyone and would be a wonderful thing we would all enjoy.

The problem is, actual driving involves too many corner cases. Turns out the human brain is really good at handling corner cases, and AI just isn’t. And even if it is as good as a human, that’s still not good enough. The combination of our litigious nature and manufacturer’s deep pockets will paint a huge target on the innovators in FSD.

I’m actually with Darthamerica and Team Human on this one. Eventually we will get there, but my money says it’s at least 5 years out and will be a triumph of the collective (with V2I communication) and not a triumph of the individual (Tesla).

Darthamerica | 19. oktober 2019

Madagascar thanks, I'm glad to find some common ground. I believe in what Tesla is doing so much that I traded a perfectly good P85+ to upgrade to a P90DL because of AP. It's the one feature that never ever ceases to amaze me. It's not just an evolution of cruise control. It's revolutionary. But it's not autonomous nor can it be with the current sensors and SW. Instead it's a highly advanced driver assistance feature. This is true of even HW3. The biggest challenge is that the SW is dependent on rules. But driving is an open loop system with near infinite variables. You can't easily code up everything. All it takes is ONE BIG CRASH to ruin everything. Humans have the ability to think and make decisions. We can make up rules as we go. That's a huge advantage. It's not unlike a video game where the AI is really good but eventually we figure it out and beat it. Yes, I know humans do all kinds of stupid things in cars and are equally dangerous. But we aren't going to ban humans from driving. This is why I think public education and redesigning roads for autonomous vehicles is needed.

Darthamerica | 19. oktober 2019

I would say we are 10 years away from Tesla's or some other car doing what Tesla shows in their promotional videos. It's my opinion that none of our current cars will ever be able to do that without strict limitations. I'd love to be wrong because it's almost time for me to get a new car.

WW_spb | 20. oktober 2019

Will be funny when Tesla gets the FSD to work next year in all of its Glory

asdfasdfa | 20. oktober 2019

adssad

jimglas | 21. oktober 2019

Have you naysayers tried using NoAP?
It drove me 3 hours home this morning without a problem
3 different freeways
I did take control in construction areas, but otherwise my car was on its own.

Darthamerica | 21. oktober 2019

@Jim I’m still on AP1 so I haven’t tried NoAP. It’s an impressive integration of features... adding GPS/NAV to lane keeping and TACC. Tesla can continue improving it even more but this is still a long way from FSD. FSD would be where you do not have to pay attention or intervene. These are the areas where the technology isn’t mature enough to handle unplanned variables that we often encounter. This is where driver education comes in. People need to understand that no matter how cool it is, for the foreseeable future there needs to be a person in the loop until HW/SW advance a bit more. My opinion is that we are a loooong way from that.

Madatgascar | 21. oktober 2019

I have used NoAP a lot. It’s pretty slick when I’m on open highways and in no hurry, but it won’t compete with aggro LA drivers for a merge in traffic. It’s probably 95% ready for the unassisted cross country trip. If it gets out of LA it will probably make it all the way across the country... until it hits the spiral merge for the Holland Tunnel.