How will Tesla demo FSD?

[Daniel in SD]

There's no difference. SDC are all about software. the reason it doesn't exist today is not because there's no hardware or difference in hardware or whether hardware is being sold to customers.

It would be impressive if there was any evidence that being able to do a self driving demo proved that one was close to building a self driving car. Plenty of companies have done self driving demos, Tesla did one themselves in 2016, but no one has a reliable self driving car yet.
The vision system is only a small part of the problem and there’s no way way to tell the difference in a demo between being 99.9% of the way there and however many 9s you need to actually make it reliable.

 

[Daniel in SD]

Thinking about it some more, if all they announce is that EAP/NoAP won't saw away at the wheel while driving around a curve, it will not only improve my user experience, but it will drive better than it ever has before. I absolutely can't stand being in cars with people that do this, and the fact my car does it on its own makes me rage. What good is a computer if it doesn't judge the curve with higher precision than a person?

That said, I still use the function every time I'm on the highway alone.

I want them to not mimick drivers who maintain speed by pulse width modulating the gas pedal. Following them makes me turn off AP.

 

[Bladerskb]

The difference is significant. Expensive lidars makes the object detection part of the solution much easier to accomplish, hence you can show progress faster to the public and the investors.

But there's a backside, either you have to adapt your hardware to be mass-producable at a reasonable cost and esthetics (nobody wants a box on the roof). Or you have to adapt your software to use cheaper hardware/cameras, which is not a simple task (if it was, why bother with the expensive hardware suite in the first place?).

Putting 1 billion dollars worth of sensors won't get you to self driving any quicker.
Its just as easy to create a demo using camera only, lidar only or camera/lidar.
having extra sensing for redundancy only matters when you already have a very good path planning system.

But to play your game here is a demo using only cameras in a normal looking car.
Is this not good enough for you. Is this not impressive to you?

 

[DrDabbles]

It would be impressive if there was any evidence that being able to do a self driving demo proved that one was close to building a self driving car. Plenty of companies have done self driving demos, Tesla did one themselves in 2016, but no one has a reliable self driving car yet.
The vision system is only a small part of the problem and there’s no way way to tell the difference in a demo between being 99.9% of the way there and however many 9s you need to actually make it reliable.

Demo or not, there's clearly a vast difference depending on where you drive. How many people in the Bay Area are ranting and raving about how navigate on autopilot is totally driving itself, Elon included. While there's obvious cognitive dissonance with some people, not all of them could be totally blind or liars.

Whatever the cause for that difference is- map resolution, higher density of trained miles, who knows. Whatever it is, it isn't present everywhere, and it's causing some people to believe FSD is mere days away while the rest of us scratch our heads wondering how anybody could be so deluded. There's an unbelievable gap between where NoAP is now, and where a truly self driving vehicle would need to be. I honestly hope they've closed that gap, but the systems architect in me...I won't believe it until I see it myself.

 

[ZeApelido]

There's no difference. SDC are all about software. the reason it doesn't exist today is not because there's no hardware or difference in hardware or whether hardware is being sold to customers.

Huh ? Better hardware allows larger networks, which generally improve accuracy

 

[Bladerskb]

So, Tesla is conducting a real driverless FSD demo and that's not good enough for you because you think it will be on a pre-mapped road like in the video (even though there is nothing in that tweet to suggest that)? Or is it that they are recording a video of it? I am confused. How are you right again?

You can't be this naive right?

The fact is that if Tesla pulls off a driverless demo of their FSD on a main road that will be damned impressive!

2012 called, they want their demo back.

funny how demos around these quarters were scoffed at now they are praised and referred to as amazing, astonishing and impressive, just because it coming from Tesla.

 

[diplomat33]

You can't be this naive right?

No, of course not. But I am also not super cynical to think that just because Tesla is recording a FSD demo that it will automatically be fake.

 

[diplomat33]

funny how demos around these quarters were scoffed at now they are praised and referred to as amazing, astonishing and impressive, just because it coming from Tesla.

Most demo videos are edited to emphasize what the company wants the public to focus on. They don't show the true state of self-driving. They only show the best. They don't show the driver disengagements and near misses. That's true for Waymo, Mobileye, Cruise, yes, including Tesla. So let's take the demos for what they are. Having said that. Demo videos can still be useful. We look to demos from Waymo, Mobileye or Cruise to see what their cars can do. Likewise, a FSD demo from Tesla should give us some insight into the current state of Tesla's FSD.

 

[Bladerskb]

Most demo videos are edited to emphasize what the company wants the public to focus on. They don't show the true state of self-driving. They only show the best. They don't show the driver disengagements and near misses. That's true for Waymo, Mobileye, Cruise, yes, including Tesla. So let's take the demos for what they are. Having said that. Demo videos can still be useful. We look to demos from Waymo, Mobileye or Cruise to see what their cars can do. Likewise, a FSD demo from Tesla should give us some insight into the current state of Tesla's FSD.

not according to everyone else the last 2 years. do you want me to compile the hundreds of scoffing posts each time a demo video was posted?

 

[diplomat33]

not according to everyone else the last 2 years. do you want me to compile the hundreds of scoffing posts each time a demo video was posted?

I don't really care about what "everyone else" is saying. If others have been inconsistent or flip flopped on their views, that's their problem.

 

[Daniel in SD]

No, of course not. But I am also not super cynical to think that just because Tesla is recording a FSD demo that it will automatically be fake.

I don’t think it will be fake either. I just think it will be exactly like the dozens of other self driving demos we’ve seen.
Maybe they’ll surprise me and release a demo of it driving 100,000 miles on a random route. Then I would believe that they’ve made real progress and aren’t just stuck in the same spot that everyone else is.

 

[BestRadar]

I want to see a Live Demo with no editing.

 

[diplomat33]

Disgusting! Short sellers were apparently tweeting about how to cause an accident with Tesla's FSD demo car.

 

[BestRadar]

Disgusting! Short sellers were apparently tweeting about how to cause an accident with Tesla's FSD demo car.

View attachment 397747View attachment 397747

That is just silly nonsense.

 

[diplomat33]

That is just silly nonsense.

It just makes the short sellers look really sad and pathetic if you ask me.

 

[DrDabbles]

Something that no FSD vehicle will have, probably ever, is intuition. Raw calculation capability can cover for a lot, and a larger array of more types of sensors can make a massive difference to how detailed the data for a scene is. But something that humans bring to the equation that computers can not is intuition.

That single fact means the types of "corner cases" that self driving vehicles will be tripped up by would be easily avoided by a human driver. Algorithmic driving may make it to the point where it's vastly superior to humans in basic cases, and pure avoidance circumstances. But if the public gets the sense that they could sense a situation is about to go bad and the car doesn't, that's going to instantly destroy confidence in the technology, and government regulators will step in.

 

[Daniel in SD]

Something that no FSD vehicle will have, probably ever, is intuition. Raw calculation capability can cover for a lot, and a larger array of more types of sensors can make a massive difference to how detailed the data for a scene is. But something that humans bring to the equation that computers can not is intuition.

That single fact means the types of "corner cases" that self driving vehicles will be tripped up by would be easily avoided by a human driver. Algorithmic driving may make it to the point where it's vastly superior to humans in basic cases, and pure avoidance circumstances. But if the public gets the sense that they could sense a situation is about to go bad and the car doesn't, that's going to instantly destroy confidence in the technology, and government regulators will step in.

They'll certainly get in very different types of accidents than humans. I'm a bit more optimistic that the public will be able to understand that they're statistically safer. After all airbags cause a certain number of deaths but the vast majority of people accept that on average they have improved safety. Part of the problem though is that every self-driving crash will be on video.

 

[diplomat33]

Something that no FSD vehicle will have, probably ever, is intuition. Raw calculation capability can cover for a lot, and a larger array of more types of sensors can make a massive difference to how detailed the data for a scene is. But something that humans bring to the equation that computers can not is intuition.

That single fact means the types of "corner cases" that self driving vehicles will be tripped up by would be easily avoided by a human driver. Algorithmic driving may make it to the point where it's vastly superior to humans in basic cases, and pure avoidance circumstances. But if the public gets the sense that they could sense a situation is about to go bad and the car doesn't, that's going to instantly destroy confidence in the technology, and government regulators will step in.

If self-driving cars cannot deal with these corner cases, the other option is to remove the corner cases so that they don't have to solve them. For example, we could change our roads to make them easier for self-driving cars to handle.

 

[Daniel in SD]

If self-driving cars cannot deal with these corner cases, the other option is to remove the corner cases so that they don't have to solve them. For example, we could change our roads to make them easier for self-driving cars to handle.

It's not the roads that are the problem, it's other road users. I bet you could have self driving today if you had only self-driving cars on the roads.

 

[diplomat33]

It's not the roads that are the problem, it's other road users. I bet you could have self driving today if you had only self-driving cars on the roads.

True. I guess I was just thinking of how some roads/intersections can be really weird, especially when lane lines are faded away or when construction forces new lanes that don't match the existing lanes. It can be unnecessarily complicated for self-driving cars. And what about when constructions workers take turns stopping traffic and cars have to merge into the incoming traffic lane before merging back into their lane. That would be very confusing for a self-driving car I would imagine since they have to read signs or hand gestures and break a lane rule by temporarily driving into incoming traffic.

But you are right that the main obstacle is other drivers. You can have the best self-driving car in the world but it's all moot if some stupid human driver decides to break the rules by running a red light, aggressively merging in front of you when it is not safe, or making an unprotected left turn in front of you when they are not supposed to. I agree 100% that if all cars on the road were self-driving, it would be a lot easier and better for all. It is other drivers which may actually prevent true L4 autonomy, not the hardware or software.