BUY FSD TODAY!

[crackers8199]

As I've said elsewhere, this comes of using the term "FSD" to mean two entirely different things! On the one hand, it's used, and commonly understood, to mean a car that does not require a driver. You can sleep in the back seat or send it by itself to pick up the kids. On the other hand, Tesla uses it, and some here use it, to refer to a package of unspecified driver-assist options which Elon optimistically believes will eventually lead to a full-self-driving car.

elon repeatedly hammered the idea of getting in your car and having it drive you to your destination with no input from you other than entering your destination. he very clearly used it as the idea of the car driving itself all the time.

 

[djgarrett21]

Elon has recently stated that FSD will be rolled out in pieces and thus will not be a binary works/doesn't situation. There is no other way to sensibly develop the functionality given Tesla's method of using us owners as beta testers.

FSD will most definitely have to be a suite of features until the time they have the hardware and software capable of full autonomous driving. But even then the capability could be very situational as the technology matures, such as; only during daytime, good weather, certain types of roads, etc.

I don't think the current hardware (sensors and computer) are capable full autonomous driving with the (eventual) level of confidence needed . Elon will come to the realization (or maybe already has) that the car will need better inputs than 8 cameras, parking sensors, and a single radar can provide. More inputs/higher resolution = higher computing power and more code.

I don't doubt Tesla is committed to delivering FSD but given the current state of AP, Tesla has a long ways to go.

 

[Drone Flyer]

I think he realizes it won't happen at all, at least for a long time due to regulation requirements. AP2 is still behind AP1. There's no way AP2 will be ready for FSD anytime soon, certainly won't be ready on any of the current Tesla cars by the time they become obsolete. There are just too many variables that remain in order for FSD to work. None of Tesla's cars can read traffic or stop signs yet. That's a major hurdle in itself that has to be overcome. I truly think that every car has to have some sort of transponder on it in order for FSD to become viable. So for this to happen, I give it 10 years before we can realistically think of autonomous driving.

 

[JeffK]

I hope I'm not offending anyone, but why would anyone pay for FSD on a two-year lease when none of the features exclusive to FSD have even been confirmed yet, much less released? Are there people who seriously thought that full self-driving was less than a year away when they bought the car?

Exactly, but some people in late 2016 actually did this and their leases are now expiring.

 

[barjohn]

My thoughts on this matter are that I want the hardware upgrade even if only pieces of FSD ever become available. I think the current hardware is inadequate for even EAP and the faster hardware may make EAP more functional. Watching the cars reaction to real time events tells me that it is struggling with the current hardware to do much predicting such that it can react more like a human to a given situation. First the hardware has to be able to decipher what its sensors are telling it. This takes more time for a computer (at present) than for the human brain. Second it must devise a strategy for dealing with the information the sensors are providing and finally it must direct the systems to react and execute the strategy while at the same time monitoring whether the system is executing the strategy as directed and is the scenario changing in real time such that the strategy must be changed and the execution halted and changed. This is a heavy computational load. Think about what it takes for the following scenario: A motorcyclist is traveling much faster than you are in traffic. He is 3 lanes over behind you. He is parallel to you can crosses 3 lanes to get ahead of you, loses control of hi bike in front of you. You don't want to hit him or the bike or the cars on either side of you and you don't want to be rear ended by the car behind you. You slam on the brakes and look to steer either left or right to avoid hitting the cyclist and at the same time avoid the cars beside you. As you try to stop and maneuver, someone else steers toward you to avoid the bike that is flying into their lane. This causes you to abort your steer toward their lane and to try to change to the opposite lane. All of this is happening is split second time. An AI system has to recognize that a motorcycle has suddenly appeared in your lane and subsequently that the motorcycle has deviated from normal motorcycle behavior and that now you have both a cycle and rider as two separate objects to contend with and avoid hitting...

I don't believe the current hardware is capable of processing information quickly enough for such a scenario. It has a hard enough time dealing with a car squeezing in to your lane and reacting at slow stop and go speeds. I won't be surprised to see level 4 capabilities require another 10x over hardware 3.0 speeds. There is a tendency to underestimate the processing the human brain does at the things it does well such as pattern and object recognition. I can hear a song I haven't heard in 50 to 60 years and suddenly the words come to me. No general purpose AI machine could recall something from such a distant past in real time that had not been called up in 50 to 60 years. It probably wouldn't have enough memory to store that much data.

For these reasons, I will probably purchase FSD in the near future.

 

[JeffK]

Watching the cars reaction to real time events tells me that it is struggling with the current hardware to do much predicting such that it can react more like a human to a given situation.

It's more than likely they simply didn't focus on making the visualization terribly pretty. The collision alarms do not go off when the virtual cars are hitting your car, so it's just a human interface issue at present.

This takes more time for a computer (at present) than for the human brain. Second it must devise a strategy for dealing with the information the sensors are providing and finally it must direct the systems to react and execute the strategy while at the same time monitoring whether the system is executing the strategy as directed and is the scenario changing in real time such that the strategy must be changed and the execution halted and changed. This is a heavy computational load.

Absolutely not true and hasn't been for a number of years. Perception with neural networks is done in linear time assuming you have the hardware to do then you'd be fine. Even in the current hardware it can detect any object in nearly 1/3 the time it takes a human being (5 ms vs 13 ms for a human). That's not even considering how long it'd take a human to react to the stimulus.

Think about what it takes for the following scenario: A motorcyclist is traveling much faster ...

Let me stop you there, if you have a hypothetical situation where even a human might struggle, think three things... First, FSD isn't crash proof. Second, if the software was there, then the computer can react faster than you can. Third, in many human situations a FSD car wouldn't get into those situations to begin with. They'll have their own set of specific FSD issues though...

It's not going to be perfect, but make no mistake the hardware is capable of perceiving and responding faster than a human. Biological neurons are slow relatively speaking.

 

[daniel]

My thoughts on this matter are that I want the hardware upgrade even if only pieces of FSD ever become available. I think the current hardware is inadequate for even EAP and the faster hardware may make EAP more functional. Watching the cars reaction to real time events tells me that it is struggling with the current hardware to do much predicting such that it can react more like a human to a given situation. First the hardware has to be able to decipher what its sensors are telling it. This takes more time for a computer (at present) than for the human brain. Second it must devise a strategy for dealing with the information the sensors are providing and finally it must direct the systems to react and execute the strategy while at the same time monitoring whether the system is executing the strategy as directed and is the scenario changing in real time such that the strategy must be changed and the execution halted and changed. This is a heavy computational load. Think about what it takes for the following scenario: A motorcyclist is traveling much faster than you are in traffic. He is 3 lanes over behind you. He is parallel to you can crosses 3 lanes to get ahead of you, loses control of hi bike in front of you. You don't want to hit him or the bike or the cars on either side of you and you don't want to be rear ended by the car behind you. You slam on the brakes and look to steer either left or right to avoid hitting the cyclist and at the same time avoid the cars beside you. As you try to stop and maneuver, someone else steers toward you to avoid the bike that is flying into their lane. This causes you to abort your steer toward their lane and to try to change to the opposite lane. All of this is happening is split second time. An AI system has to recognize that a motorcycle has suddenly appeared in your lane and subsequently that the motorcycle has deviated from normal motorcycle behavior and that now you have both a cycle and rider as two separate objects to contend with and avoid hitting...

I don't believe the current hardware is capable of processing information quickly enough for such a scenario. It has a hard enough time dealing with a car squeezing in to your lane and reacting at slow stop and go speeds. I won't be surprised to see level 4 capabilities require another 10x over hardware 3.0 speeds. There is a tendency to underestimate the processing the human brain does at the things it does well such as pattern and object recognition. I can hear a song I haven't heard in 50 to 60 years and suddenly the words come to me. No general purpose AI machine could recall something from such a distant past in real time that had not been called up in 50 to 60 years. It probably wouldn't have enough memory to store that much data.

For these reasons, I will probably purchase FSD in the near future.

In the scenario you describe, 99% of human drivers would end up crashing, either into the motorcycle or one of the other cars. I doubt that EAP could handle it today either. And there are situations in which no system could prevent a crash. The point is: At what stage does the computer make you safer than you are without it? I think we are there already. I think that EAP makes me safer. With the proviso that EAP assists me. It does not take over responsibility from me.

I agree with you that full self-driving will require more sensors and more computing power than my car has. I won't speculate on whether HW 3 provides enough computing power. But I don't think the Model 3's sensors are adequate.

 

[barjohn]

@JeffK, detection is not the same thing as recognition. Recognizing that a detected object is a motorcycle,especially when it is at an angle to the cameras is not likely happening faster than the human brain. Recognizing that it has crashed and that what is being detected is a motorcycle spinning around at all kinds of different angles takes a lot more computing and thus time. If the NN has never seen a motorcycle crash and learned that is what a motorcycle crash looks like it would take it even longer. The deeper the NN the longer it takes.

 

[JeffK]

@JeffK, detection is not the same thing as recognition. Recognizing that a detected object is a motorcycle,especially when it is at an angle to the cameras is not likely happening faster than the human brain. Recognizing that it has crashed and that what is being detected is a motorcycle spinning around at all kinds of different angles takes a lot more computing and thus time. If the NN has never seen a motorcycle crash and learned that is what a motorcycle crash looks like it would take it even longer. The deeper the NN the longer it takes.

Detection that it is a motorcycle happens far faster than a human. We won't talk about "recognition" in the context of NNs because that's a loaded word. As far as the depth of the NN, it's irrelevant if the hardware supports running in in real time with the incoming frames. Your limitation at that point is the cameras.

To be clear, a frame takes the same amount of time to process through any set NN no matter what it contains. Just because the scene changes doesn't mean that seen takes longer to process as far as image recognition is concerned.

 

[MXWing]

What people really fail to realize here is that just because OUR Tesla’s can not read signs and lights, drive on Nav, auto lane change, does not mean TESLA builds cannot.

With the propensity for dumb humans to find the most extreme of corner cases that will kill them, Tesla has to be ultra conservative.

The fleet has to detect signs across billions of shadow miles before they can even think of putting it into early access.

If Tesla is already running mad max, drive on nav, sign recognition it’s coming to us.

Question is not IF but when.

If my Model X/Model 3 can’t be upgraded past full Level 3 I can live with that.

I paid 3-5K for ‘FSD’ not 30,000-50,000.

They can call it bio defense mode, they can call it autopilot, full self drive, etc and it’s not wholly accurate. It’s sexing it up aka marketing. Consumers need to be more discerning.

My Volt has ‘lane keeping assist’. Assist my ass. It would be like my preschooler helping me ‘build a computer’. They are not assisting, they create more work for me.

 

[Daniel in SD]

Audi is not stupid and did not approve it in California. They know it would mean full autonomous driving in the entire LA county all day long.

Wouldn’t that be the point? They’re not selling it yet but I’m sure they’ll apply to get it approved in CA. I’m sure it’s their biggest market in the US.

 

[daniel]

@JeffK, detection is not the same thing as recognition. Recognizing that a detected object is a motorcycle,especially when it is at an angle to the cameras is not likely happening faster than the human brain. Recognizing that it has crashed and that what is being detected is a motorcycle spinning around at all kinds of different angles takes a lot more computing and thus time. If the NN has never seen a motorcycle crash and learned that is what a motorcycle crash looks like it would take it even longer. The deeper the NN the longer it takes.

The computer does not need to recognize that it's a motorcycle. It just has to react to the threat of a collision in such a way as to avoid the collision more often than a human would. Because that's the bottom line: We'll never achieve zero fatal accidents. The goal of self-driving cars is, first, to be safer than human drivers, and ultimately, to get the accident rate down as low as we can.

EAP plus an alert driver is already safer than a driver alone. I don't think the Model 3 has the hardware needed to be safer driverless than EAP plus an alert driver. But maybe the next generation Tesla will. And I fully expect that it's the future. In my lifetime the first artificial satellite was put into space, the first unbreakable shampoo bottle went on sale, we learned that continents drift and that there are galaxies beyond our own, humans walked on the moon, transistor radios came to the consumer market, the first handheld electronic calculators went on the market, desktop computers, cell phones, smartphones, and tablet computers appeared, and now I have a car with EAP that uses no gasoline and gets its energy from hydro. Maybe I won't live to see a fully-driverless car. Maybe I will.

 

[schonelucht]

It really bites people who've bought FSD and were on a two year lease.

All Two year lease offers were for AP1 cars if I remember correctly. But the point might still stand by the time the first three year leases roll around.

 

[mhan00]

What people really fail to realize here is that just because OUR Tesla’s can not read signs and lights, drive on Nav, auto lane change, does not mean TESLA builds cannot.

With the propensity for dumb humans to find the most extreme of corner cases that will kill them, Tesla has to be ultra conservative.

The fleet has to detect signs across billions of shadow miles before they can even think of putting it into early access.

If Tesla is already running mad max, drive on nav, sign recognition it’s coming to us.

Question is not IF but when.

If my Model X/Model 3 can’t be upgraded past full Level 3 I can live with that.

I paid 3-5K for ‘FSD’ not 30,000-50,000.

They can call it bio defense mode, they can call it autopilot, full self drive, etc and it’s not wholly accurate. It’s sexing it up aka marketing. Consumers need to be more discerning.

My Volt has ‘lane keeping assist’. Assist my ass. It would be like my preschooler helping me ‘build a computer’. They are not assisting, they create more work for me.

Yeah, the Volt lane keeping assist is useless. I was super excited to try it out when I first got the car two and a half years back, and when I did (late at night on an empty road because I value my life and that of others) it went over the line about halfway before ping ponging back and doing the same thing on the other side. The TACC, on the other hand, was a “mother of god where has this been all my life?” moment. AP, so far, has been a step or two above that. I may re-evaluate when I experience the phantom braking/other issues other people have encountered, but so far so good when I’ve been on the highway.

 

[Garlan Garner]

I never claimed there was breach of promise on EAP. I said that if Tesla does not deliver on its promise of full self-driving coming to the owners of Model 3s who paid for the FSD package, it would be breach of promise.

And to date there are no "FSD features." Buyers of the FSD package have not yet received any features that buyers of EAP have not also gotten. And while Tesla has backtracked from promising a full self-driving car, to talking about the "FSD" options package, they were talking about cars that do not require anybody in the driver's seat when they were selling the FSD package. People bought the FSD package because Tesla told them that at some unspecified date in the future their car would not require a driver. If that date comes after the reasonable expected lifetime of the car, that would be breach of promise.

Tesla overpromised, and that could bite them in the ass. I never believed that true full self-driving would come before it's time for me to trade up to a newer car, or that the sensors in the 2018 Model 3 would be adequate for full self-driving, so I bought EAP but not FSD. FSD is a pig in a poke at this time. EAP is fabulous and I love my car.

IMO "Drive-On-Nav" is more FSD feature than EAP.

All they have to do is re-allocate it. Then FSD would have a feature.

My point - FSD does NOT have to be fully rolled out for FSD to exist. EAP is not fully deployed and people are satisfied.


Watch the poll.....don't just trust me.
Do you regret buying (or not buying) FSD?

 

[Garlan Garner]

we only have features of EAP because it's a suite of features. you seem to keep ignoring that the other feature is named "FULL SELF DRIVING." either the car drives itself or it doesn't. if it doesn't, then it doesn't fit the term "full self driving."

if they never intended FSD to be the car driving itself all the time, they should have named it something different.

Absolutely not. Enhanced Auto Pilot "isn't" an autopilot.

Its a drivers assist group of features so far. Its not auto piloting because it can't make real "pilots" decisions yet. Far from it.
You know why people aren't complaining about EAP not being a full out autopilot at this moment?

 

[Garlan Garner]

I hope I'm not offending anyone, but why would anyone pay for FSD on a two-year lease when none of the features exclusive to FSD have even been confirmed yet, much less released? Are there people who seriously thought that full self-driving was less than a year away when they bought the car?

I put a deposit on a car "sight unseen" - 2.7 years before I got it.

I trusted TESLA ( and I was right ).

I purchased FSD and it has actually paid off as far as cost is concerned.




1year from now I'll make the same statement about FSD.

I trusted TESLA ( and I was right ).

Watch the poll.....don't just trust me.
Do you regret buying (or not buying) FSD?​

 

[cgalpin]

I paid for FSD on my M3 but now on Tesla.com it is gone. It only shows EAP.

Same here. Interesting.

Didn't Elon already say that those who paid for FSD would get the new chip 6 months before everyone else?

I paid for FSD figuring at a minimum i would get the opportunity for FSD "features" sooner than waiting to get it all outright, if that ever occurs. Think of it as paying to be a beta tester if you want, but count me in!

Don't worry, I bought stock too

 

[Garlan Garner]

Same here. Interesting.

Didn't Elon already say that those who paid for FSD would get the new chip 6 months before everyone else?

I paid for FSD figuring at a minimum i would get the opportunity for FSD "features" sooner than waiting to get it all outright, if that ever occurs. Think of it as paying to be a beta tester if you want, but count me in!

Don't worry, I bought stock too

exactly

Do you regret buying (or not buying) FSD?​

 

[cgalpin]

exactly

Do you regret buying (or not buying) FSD?​

Definitely do not regret buying it.