Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Seems like FSD is a complete crock

This site may earn commission on affiliate links.
Thank you! According to this informative graph, we will be there in just 10 (or more) short years! And that is for Level 4, which will apparently spend substantially more time in the nice comfortable trough. Somehow level 5 is not even at the peak of expectations.
10+ short years for level 4 vehicles to have reached "plateau of productivity" when they will be ubiquitous (i.e. reached economies of scale). May be 10% of the cars on the road would be at L4.

Arguably L4 AVs are here now but they just need to scale up. Given the slow spread of any new technology in cars (the Toyota we bought recently has "entune" that definitely looks worse than iPhone 1.0) - Gartner is right to put that as 10+ years.

Funny twist in this is - if Tesla gets to L4 (with current sensor suit) - instantly we'll have millions of cars on the road that are L4. By 2025, Tesla might have sold 10M cars.

Anyway - my point was wherever you read - you can see this trough of despair w.r.t. FSD. Yet, we are definitely at a better place now than 3 years back. i.e. don't give into the hype cycle but rationally analyze the situation with the information we have.
 
Last edited:
I like how trendy this type of graph is these days. I’ve seen it in a few places but not sure it has any scientific basis.

Anyway, I can see level 4 relatively soon (3-4 years) since it can be done in a limited domain. And then progressively more areas as they program in all the special case intersections or whatever they have to do for that tricky stuff.

Level 5 seems super hard though. Just so many possibilities. Not convinced that the computer brain with 8 eyes will be able to compete with a human brain for quite some time. This isn't like playing chess or Go. I will be really impressed if it happens in 5-10 years, but I hope I'm wrong.
 
Does the car even autopark?

Does adaptive cruise control work well or phantom break a lot?

I don’t care much about the car. I have cars. I want the tech Elon is selling. I think he’s duped me into wanting this thing.
I drove on Navigate on AP between DC and NYC this weekend, 99% of it was NOA. I think it phantom breaked once. Who cares? I could basically go to sleep and the car would drive itself. It's amazing.
 
  • Like
Reactions: diplomat33
I drove on Navigate on AP between DC and NYC this weekend, 99% of it was NOA. I think it phantom breaked once. Who cares? I could basically go to sleep and the car would drive itself. It's amazing.

Just have to hope there's no road debris and no deer on your route.

We're in a dangerous time right now. AP is very, very good, and NoA is getting to be adequate - so it's easy to trust the car to handle everything and lose your situational awareness. But the car won't do anything about a pothole, an object on the road, or a deer - so if you aren't paying attention and one of them is in your lane, it'll get hit.

I'm not ready to endorse the theory that less mature systems are safer because folks have to pay attention to them, but I do think Tesla needs to move to Level 3 as quickly as they safely can, so the cars will alert for or react to the debris and potholes and deer.
 
Level 3 as quickly as they safely can, so the cars will alert for or react to the debris and potholes and deer.

I have a hard time comprehending how the system will be good enough to do this any time soon. It’s difficult enough for a human to distinguish marks on the road from road debris, as it is. I could see partially effective animal identification soon. But relatively small road debris and potholes...going to be a long, long time. Probably will be more effective for the next 5-10 years to constantly be using crowd-sourced data and alert driver to intervene to avoid upcoming significant hazards (or have system go to hyper-aware “jumpy” mode - probably not the best).
 
I'm not ready to endorse the theory that less mature systems are safer because folks have to pay attention to them, but I do think Tesla needs to move to Level 3 as quickly as they safely can, so the cars will alert for or react to the debris and potholes and deer.

I don't subscribe to that theory either. We should press forward to an autonomous system (L4/5) where the driver does not need to pay attention, not regress to a lesser one just so that the driver has to pay attention more. I think that is what you want too.

But I think in these instances, L4 would be better than L3. For one, if your car is capable of warning you about debris, a pothole or a deer, then it should be able to take action too. And isn't it better for the car to just avoid these hazards directly rather than warn the driver and hope the driver takes action? Cut out the middle man! Also, L3 requires a 10 second warning to give an inattentive driver enough time to get back their situational awareness and take action. In the case of a deer, you don't have 10 seconds to act because it can happen so quickly. So L3 won't work to avoid deer. So L4 is better: just get the car to avoid these hazards directly.

So really the moral of the story is that Tesla needs to hurry up finishing their vision NN because then and only then, will Autopilot be able to be full self-driving and take action without needing the driver to pay attention.
 
I have a hard time comprehending how the system will be good enough to do this any time soon. It’s difficult enough for a human to distinguish marks on the road from road debris, as it is. I could see partially effective animal identification soon. But relatively small road debris and potholes...going to be a long, long time. Probably will be more effective for the next 5-10 years to constantly be using crowd-sourced data and alert driver to intervene to avoid upcoming significant hazards (or have system go to hyper-aware “jumpy” mode - probably not the best).

I like the idea of an integrated, automated Waze type service. That's been suggested before - though not in the context of pre-alerting the driver under level 3 based on it. That's an interesting idea.

If the car knows which lane(s) are affected from prior cars, NoA can just switch away from them and keep going in peace.
 
So really the moral of the story is that Tesla needs to hurry up finishing their vision NN because then and only then, will Autopilot be able to be full self-driving and take action without needing the driver to pay attention.

Sure. I just find it inconceivable that this incredibly hard problem will be solved by Tesla or anyone else any time soon. I hope I'm wrong. Just find it hard to believe that such a system will be as good as a hyper-aware human driver any time soon for such edge cases (which are quite common). This is the reason a lot of people use Waze as a warning system - because it is SO difficult to react in a timely manner to arbitrary objects unless you know they're coming and are in hyper-aware avoidance mode. I understand reaction time of a system will be much faster than a human. But perception seems really, really difficult.
 
Sure. I just find it inconceivable that this incredibly hard problem will be solved by Tesla or anyone else any time soon. I hope I'm wrong. Just find it hard to believe that such a system will be as good as a hyper-aware human driver any time soon for such edge cases (which are quite common). This is the reason a lot of people use Waze as a warning system - because it is SO difficult to react in a timely manner to arbitrary objects unless you know they're coming and are in hyper-aware avoidance mode. I understand reaction time of a system will be much faster than a human. But perception seems really, really difficult.

Perception is hard but it is doable. It just requires "grinding" through machine learning: you collect a ton of images, then annotate the images by hand and feed them into your computer until it "learns" what the images are. This works with everything. You can collect a lot of images of potholes and feed them into the machine and it will learn to recognize potholes. You can collect a lot of images of various road debris and teach the machine to recognize debris. You can use this method for literally everything, debris, potholes, deer, road markings etc... It's what Tesla is doing but it takes time. In fact, Tesla has already used this method successfully with lane lines, cars, trucks, pedestrians, cyclists, traffic lights and probably more. It's just not finished yet.
 
you collect a ton of images, then annotate the images by hand and feed them into your computer until it "learns" what the images are. This works with everything. You can collect a lot of images of potholes and feed them into the machine and it will learn to recognize potholes. You can collect a lot of images of various road debris and teach the machine to recognize debris. You can use this method for literally everything, debris, potholes, deer, road markings etc... It's what Tesla is doing but it takes time.

The thing is that no one knows whether this will actually work with high enough sensitivity and specificity. As I said, I believe it can work for animals and such, most of the time. However, for road debris I'm not convinced, given the massive array of shapes and sizes of debris that match pre-existing harmless marks, that exist all over the road today.

To be clear, I'm not convinced there is any set of sensors out there that will really be good enough in the near term. Obviously in an abstract sense, it is possible.
 
The thing is that no one knows whether this will actually work with high enough sensitivity and specificity. As I said, I believe it can work for animals and such, most of the time. However, for road debris I'm not convinced, given the massive array of shapes and sizes of debris that match pre-existing harmless marks, that exist all over the road today.

To be clear, I'm not convinced there is any set of sensors out there that will really be good enough in the near term. Obviously in an abstract sense, it is possible.

Whitelist? Use all of the human drivers to map the innocent markings and assemble them into downloadable tiles for FSD...

That's a brute force solution that'll involve lots of individual listings, but then if you find something not on the list, alert for it or dodge it, and if the car can see it's flat when the car gets close enough, then add it to the list?