FSD Level 5

[lalt3]

Hey guys,

I was curious if anyone has any information regarding full autonomy, specifically level 5 FSD.

Obviously it will take some time to hit this accolade, but when it does happen, will Tesla make their programs open source?

I’m envisioning something where an address gets inputted, and the vehicle drives itself there, and returns back, basically like a robo taxi, but for companies apart from Tesla.

For example Uber; Say a passenger enters the pickup/dropoff location in the Uber app, the app tells the Tesla the location, the tesla drives itself there, picks up, and drops off.

I would love to hear everyone’s thoughts on this!

 

[JulienW]

Hey guys,

I was curious if anyone has any information regarding full autonomy, specifically level 5 FSD.....

Two weeks

 

[mtndrew1]

Hey guys,

I was curious if anyone has any information regarding full autonomy, specifically level 5 FSD.

Obviously it will take some time to hit this accolade, but when it does happen, will Tesla make their programs open source?

I’m envisioning something where an address gets inputted, and the vehicle drives itself there, and returns back, basically like a robo taxi, but for companies apart from Tesla.

For example Uber; Say a passenger enters the pickup/dropoff location in the Uber app, the app tells the Tesla the location, the tesla drives itself there, picks up, and drops off.

I would love to hear everyone’s thoughts on this!

Oh you sweet precious flower child…

 

[diplomat33]

Hey guys,

I was curious if anyone has any information regarding full autonomy, specifically level 5 FSD.

Obviously it will take some time to hit this accolade, but when it does happen, will Tesla make their programs open source?

I’m envisioning something where an address gets inputted, and the vehicle drives itself there, and returns back, basically like a robo taxi, but for companies apart from Tesla.

For example Uber; Say a passenger enters the pickup/dropoff location in the Uber app, the app tells the Tesla the location, the tesla drives itself there, picks up, and drops off.

I would love to hear everyone’s thoughts on this!

L5 autonomy means the car can drive itself unsupervised on all paved roads, anywhere and anytime a typical human driver can drive. So L5 in the US would mean the car would drive itself completely unsupervised on all paved roads in the entire US, all weather, day and night. In the event where driving becomes impossible, like a white out blizzard, the car automatically pulls over until conditions improve. IMO, L5 is not going to happen for awhile. Tesla is certainly far from L5. Teslas are not going to be robotaxis where you just input an address anywhere in the Uber app and the Tesla picks you up.

And there are companies like Waymo that already have robotaxis where you just input an address and the driverless car picks you up. But they are only available in select areas so far. Nobody has robotaxis that are available everywhere yet.

 

[johnny_cakes]

I dream that one day my car will shuttle me to the top of a mountain bike trail and pick me up at the bottom while I rip down the trail with my bike. Oh it's a beautiful dream indeed. But based on my experience with FSD Beta today, I'd settle for it to just be able to cross a 4 way stop without embarrassing me and annoying everyone behind me.

 

[kabin]

It sounds like all they need to do is sit back and things becomes easier than they originally thought.

I agree with two weeks.

 

[RTPEV]

This is so far in the future, anything could happen by then...

But why on earth would Tesla open source their code?

The reason why it would be open sourced at all would be during the development phase because those with the vision for what it could become did not have the resources to go it alone and would open it up to the development community to accelerate development.

This is not what is happening with FSD. Tesla most certainly has the resources to go it alone and in fact they see it as a huge value. What they HAVE talked about doing, however, is licensing it to third parties (i.e. other automakers) to put into their vehicles.

 

[E90alex]

**IF** Tesla ever gets there, they would just create their own ride share service. Why share profit with a different company.

 

[lalt3]

L5 autonomy means the car can drive itself unsupervised on all paved roads, anywhere and anytime a typical human driver can drive. So L5 in the US would mean the car would drive itself completely unsupervised on all paved roads in the entire US, all weather, day and night. In the event where driving becomes impossible, like a white out blizzard, the car automatically pulls over until conditions improve. IMO, L5 is not going to happen for awhile. Tesla is certainly far from L5. Teslas are not going to be robotaxis where you just input an address anywhere in the Uber app and the Tesla picks you up.

And there are companies like Waymo that already have robotaxis where you just input an address and the driverless car picks you up. But they are only available in select areas so far. Nobody has robotaxis that are available everywhere yet.

I dream that one day my car will shuttle me to the top of a mountain bike trail and pick me up at the bottom while I rip down the trail with my bike. Oh it's a beautiful dream indeed. But based on my experience with FSD Beta today, I'd settle for it to just be able to cross a 4 way stop without embarrassing me and annoying everyone behind me.

From what I’ve read, right now FSD improvements are hindered not due to engineering constraints, but because of computing power. Looks like they have their sights on the new Dojo super computer to help with that.

We’ve never seen Tesla with this type of processing power, so I’m wondering if Elons latest prediction of Lvl 4/5 FSD by later this year will actually prevail.

 

[JulienW]

From what I’ve read, right now FSD improvements are hindered not due to engineering constraints, but because of computing power. Looks like they have their sights on the new Dojo super computer to help with that.

We’ve never seen Tesla with this type of processing power, so I’m wondering if Elons latest prediction of Lvl 4/5 FSD by later this year will actually prevail.

Dojo to the L5 rescue.

 

[lalt3]

This is so far in the future, anything could happen by then...

But why on earth would Tesla open source their code?

The reason why it would be open sourced at all would be during the development phase because those with the vision for what it could become did not have the resources to go it alone and would open it up to the development community to accelerate development.

This is not what is happening with FSD. Tesla most certainly has the resources to go it alone and in fact they see it as a huge value. What they HAVE talked about doing, however, is licensing it to third parties (i.e. other automakers) to put into their vehicles.

I wasn’t saying license it completely, obviously they aren’t going to give away their technology for free.

What I am getting at, however, is letting other companies with fleets of Teslas utilize FSD to scale their own operations.

For example a shipping company with a fleet of Tesla Semi’s being able to create/edit their own logistics management software to allow trucks to autonomously drive to and from the destination once the cargo has been loaded/unloaded. The software would really only need the ability to input information into the FSD System / GPS. Do you see this as a potential option?

 

[lalt3]

L5 autonomy means the car can drive itself unsupervised on all paved roads, anywhere and anytime a typical human driver can drive. So L5 in the US would mean the car would drive itself completely unsupervised on all paved roads in the entire US, all weather, day and night. In the event where driving becomes impossible, like a white out blizzard, the car automatically pulls over until conditions improve. IMO, L5 is not going to happen for awhile. Tesla is certainly far from L5. Teslas are not going to be robotaxis where you just input an address anywhere in the Uber app and the Tesla picks you up.

And there are companies like Waymo that already have robotaxis where you just input an address and the driverless car picks you up. But they are only available in select areas so far. Nobody has robotaxis that are available everywhere yet.

Sorry forgot to quote you, I’ll copy and paste my previous messages so you see what I mean. Would love to your feedback-

“From what I’ve read, right now FSD improvements are hindered not due to engineering constraints, but because of computing power. Looks like they have their sights on the new Dojo super computer to help with that.

We’ve never seen Tesla with this type of processing power, so I’m wondering if Elons latest prediction of Lvl 4/5 FSD by later this year will actually prevail.

I wasn’t saying license it completely, obviously they aren’t going to give away their technology for free.

What I am getting at, however, is letting other companies with fleets of Teslas utilize FSD to scale their own operations.

For example a shipping company with a fleet of Tesla Semi’s being able to create/edit their own logistics management software to allow trucks to autonomously drive to and from the destination once the cargo has been loaded/unloaded. The software would really only need the ability to input information into the FSD System / GPS. Do you see this as a potential option?“

 

[JulienW]

So OP are you an original FSD Beta (10.2) tester?

There is just no way our cars will have the OEDR or ADS for L5 and I suspect we will never even see a limited ODD with L4. As an original tester the DDT is no where close to L3 and I doubt we will ever have a full ODD L3 system. Tesla could maybe have a limited ODD (like highway) L3 system but I think we are more likely stuck with a L2 "work in progress" system that improves but is always "two weeks" away.

I myself would be tickled pink with a full ODD L3 system but based on what I drive/experience I have hard time seeing that happen.

 

[fasteddie7]

I switched hardware for a few days (drove HW3 around instead of 4) and I can't currently envision how it's going to achieve this when any slight change in weather causes the cameras to throw up a FSD degraded message. Couple this with the general population being jerks (I think of the traffic cones on Chevy Cruze) where the same folks that coal roll and block superchargers on purpose come up with some way to cover cameras or place an item that causes the vehicle to get "stuck", (I smell a tiktok trend) and it could be a a much longer time than originally thought as after the roadway problems are solved, (if ever) engineers have the challenge to solve for people's stupidity. The promise has been talked about for someone at this point that I've watched the joke "two weeks" become the new "three months maybe, six months definitely"

 

[uscbucsfan]

I switched hardware for a few days (drove HW3 around instead of 4) and I can't currently envision how it's going to achieve this when any slight change in weather causes the cameras to throw up a FSD degraded message. Couple this with the general population being jerks (I think of the traffic cones on Chevy Cruze) where the same folks that coal roll and block superchargers on purpose come up with some way to cover cameras or place an item that causes the vehicle to get "stuck", (I smell a tiktok trend) and it could be a a much longer time than originally thought as after the roadway problems are solved, (if ever) engineers have the challenge to solve for people's stupidity. The promise has been talked about for someone at this point that I've watched the joke "two weeks" become the new "three months maybe, six months definitely"

I will say that FSD has made leaps and bounds when there is adverse weather. It used to disable for bad weather. Now it just displays FSD degraded, but most of the time it works fine. The only annoying part is when it limits your speed when the weather isn't bad. I get it if there is 0 visibility, but I've had the rain stop and THEN FSD locked my speed at 10 under the speed limit.

 

[JB47394]

I’m wondering if Elons latest prediction of Lvl 4/5 FSD by later this year will actually prevail.

If a statement by a CEO cannot be verified, ignore it. It's just a CEO painting a rosy picture about their company. My assumption is that I'll continue to monitor my car as it drives around. If I can do that with only monitoring of where I'm looking, I'll consider that a big win. If Tesla figures out how to do more, great, but I'll immediately ignore anyone who speaks of the future of research projects. Vehicle autonomy? Fusion power? Show me.

For example a shipping company with a fleet of Tesla Semi’s being able to create/edit their own logistics management software to allow trucks to autonomously drive to and from the destination once the cargo has been loaded/unloaded. The software would really only need the ability to input information into the FSD System / GPS.

An app system would provide the functionality that you're after. Just as our cell phones capabilities are enhanced by loading apps, so too could a Tesla vehicle. Those apps would provide functionality specific to a given company's needs, but drawing on the abilities of their Tesla vehicles. There's no need to go open source for that.

 

[fasteddie7]

I will say that FSD has made leaps and bounds when there is adverse weather. It used to disable for bad weather. Now it just displays FSD degraded, but most of the time it works fine. The only annoying part is when it limits your speed when the weather isn't bad. I get it if there is 0 visibility, but I've had the rain stop and THEN FSD locked my speed at 10 under the speed limit.

one has to assume engineers are hard at work solving for all these issues, and months from now we will all be saying how much better it is than it was when we had these discussions, the barrier I see is the one that engineers don’t see coming because the stupidity of people has no bounds. they are going to have to account for whatever the people who currently light firecrackers in their pants, jump off rooftops onto folding tables, and stuff spoonfuls of dry powder into their mouths are going to do to “mess up“ a driverless Tesla when they see it for likes and views. I can see it now, the app pops up a message telling us your Tesla has been delayed due to person trying to joust with it in a shopping cart wearing a cape and dunce cap eating green jello from a conch shell.

 

[diplomat33]

From what I’ve read, right now FSD improvements are hindered not due to engineering constraints, but because of computing power. Looks like they have their sights on the new Dojo super computer to help with that.

We’ve never seen Tesla with this type of processing power, so I’m wondering if Elons latest prediction of Lvl 4/5 FSD by later this year will actually prevail.

That's certainly Elon's talking point. The idea is that solving L5 simply requires training big enough neural nets. And to train neural nets big enough in a reasonable amount of time requires massive computing power. Ergo, with big enough computing power like Dojo, solving L5 should just be a matter of time. I think Elon is probably right in theory but completely wrong on how long it will take.

I think Dojo will help Tesla make a lot of progress with FSD. But I am skeptical of Elon's timeline of solving FSD by end of this year. The fact is that every leader in FSD uses massive computing power to training large neural nets. Tesla is not doing anything unique with Dojo. Other companies have supercomputers bigger than Dojo. For example, Waymo has all the power of Google supercomputers which have more computing power than Dojo. Baidu (the Chinese Google) has supercomputers with more computing power than Dojo. Both Waymo and Baidu have deployed driverless robotaxis more capable than FSD beta. Waymo is poised to have driverless robotaxis in 4 cities by the end of this year! Yet, with all that compute and actual driverless robotaxis, Waymo and Baidu have not solved L5 yet. Simply having massive computing power does not automatically solve FSD. It also matters how you use that computing power. You need the right data and the right training to solve FSD. Having massive computing power just makes the training go faster, which is very important, but is not the only requirement.

As I said in another thread, Elon's prediction of L4 or L5 makes no sense. L5 is not more reliable L4. So it is not a matter of FSD achieving L4 or L5 depending on how reliable it is. L4 and L5 have completely different ODDs. L4 can have any limited ODD. For example, L4 can have an ODD as small as a 10 sq mi geofence or 99% of US roads except bad weather. L5 requires an "unlimited" ODD, defined as all paved roads and all conditions that a typical human driver can handle. So I don't think we should mix L4/L5. IMO, Tesla could certainly do L4 by end of this year, if the ODD was small enough. But L5 is very different. I don't think Tesla will achieve L5 by end of this year. I am not saying that Tesla will never solve L5, I just think it will take longer than Elon thinks. I don't think it will happen by end of this year. Tesla told the CA DMV that removing driver supervision requires an intervention rate of 1-2M miles per intervention! So Tesla would need to achieve 1-2M miles per intervention everywhere in the entire US, to achieve L5 in the US. Tesla is nowhere close to that. Will Dojo allow Tesla to go from the current intervention rate to 1-2M miles per intervention in just the next 4-5 months? Unlikely IMO.

Conclusion: Will Dojo allow Tesla to make lots of progress with FSD? Yes, absolutely. Will Dojo magically solve L5 by end of this year? Unlikely IMO.

 

[jeewee3000]

one has to assume engineers are hard at work solving for all these issues, and months from now we will all be saying how much better it is than it was when we had these discussions, the barrier I see is the one that engineers don’t see coming because the stupidity of people has no bounds. they are going to have to account for whatever the people who currently light firecrackers in their pants, jump off rooftops onto folding tables, and stuff spoonfuls of dry powder into their mouths are going to do to “mess up“ a driverless Tesla when they see it for likes and views. I can see it now, the app pops up a message telling us your Tesla has been delayed due to person trying to joust with it in a shopping cart wearing a cape and dunce cap eating green jello from a conch shell.

This argument has been made before, but it shouldn't hinder Tesla Vision.
(For example many state: what if you block the road/paint weird road markings/show a picture of a stop sign etc to a self driving vehicle).

The thing is that actions like this are crimes and/or vandalism.

Also: right now people aren't sabotaging the road signs/markings to screw up manually driven cars, whilst there are many possibilities to do so.

Therefore I see the whole "dumb people will cover up camera's" argument as only a minor occurance, like vandalism. (People thought charging your car in a public place would immediately result in vandalism to the charger/cable/car but these fears have not panned out. There are cases of vandalism but these are as rare as with vandalims on ICE cars).

TL;DR: this is the least of our worries. The main question is: can vision + machine learning pull it off? George Hotz seems to think so, but said "it could be in 1 year, 10 years or 30 years". (Latest George Hotz interview by Lex Fridman, a worthwile listen. He is very confident that Tesla FSD is on the right path and they know exactly how to achieve FSD (which architecture), but that they just have to now build it and train it.)

 

[jeewee3000]

That's certainly Elon's talking point. The idea is that solving L5 simply requires training big enough neural nets. And to train neural nets big enough in a reasonable amount of time requires massive computing power. Ergo, with big enough computing power like Dojo, solving L5 should just be a matter of time. I think Elon is probably right in theory but completely wrong on how long it will take.

I think Dojo will help Tesla make a lot of progress with FSD. But I am skeptical of Elon's timeline of solving FSD by end of this year. The fact is that every leader in FSD uses massive computing power to training large neural nets. Tesla is not doing anything unique with Dojo. Other companies have supercomputers bigger than Dojo. For example, Waymo has all the power of Google supercomputers which have more computing power than Dojo. Baidu (the Chinese Google) has supercomputers with more computing power than Dojo. Both Waymo and Baidu have deployed driverless robotaxis more capable than FSD beta. Waymo is poised to have driverless robotaxis in 4 cities by the end of this year! Yet, with all that compute and actual driverless robotaxis, Waymo and Baidu have not solved L5 yet. Simply having massive computing power does not automatically solve FSD. It also matters how you use that computing power. You need the right data and the right training to solve FSD. Having massive computing power just makes the training go faster, which is very important, but is not the only requirement.

As I said in another thread, Elon's prediction of L4 or L5 makes no sense. L5 is not more reliable L4. So it is not a matter of FSD achieving L4 or L5 depending on how reliable it is. L4 and L5 have completely different ODDs. L4 can have any limited ODD. For example, L4 can have an ODD as small as a 10 sq mi geofence or 99% of US roads except bad weather. L5 requires an "unlimited" ODD, defined as all paved roads and all conditions that a typical human driver can handle. So I don't think we should mix L4/L5. IMO, Tesla could certainly do L4 by end of this year, if the ODD was small enough. But L5 is very different. I don't think Tesla will achieve L5 by end of this year. I am not saying that Tesla will never solve L5, I just think it will take longer than Elon thinks. I don't think it will happen by end of this year. Tesla told the CA DMV that removing driver supervision requires an intervention rate of 1-2M miles per intervention! So Tesla would need to achieve 1-2M miles per intervention everywhere in the entire US, to achieve L5 in the US. Tesla is nowhere close to that. Will Dojo allow Tesla to go from the current intervention rate to 1-2M miles per intervention in just the next 4-5 months? Unlikely IMO.

Conclusion: Will Dojo allow Tesla to make lots of progress with FSD? Yes, absolutely. Will Dojo magically solve L5 by end of this year? Unlikely IMO.

Yup.

FSD will see a "chatGPT-moment" where it suddenly works much better than people ever thought it would/could do, possibly within the next few years.

However, like with ChatGPT3 (which brought ChatGPT in the media, and brought ChatGPT 1 and 2 to shame), people will immediately after realise that - even though the newest iteration is a great step forward - a lot of work lies ahead for true autonomy.

Some (like Omar) seem to already think this ChatGPT-moment has happened. I do not. So true L5 will be some years after the "FSD-ChatGPT-moment".

Let's say 2 years for the FSD ChatGPT moment, and then 3 years after that for L5, I'm leaning towards 2028 for Tesla Vision L5. (Most likely HW5 or 6).

As an investor I just hope there will be more/grander profits from FSD before that time.