Elon: Cybertruck's air suspension will be linked to FSD and respond automatically

[EVCanuck]

If a body (e.g. a human body) is to be moved along a curved path of radius r it must have a force m*r*omega_dot in which m is its mass and omega_dot its angular velocity (speed divided by r) acting on it perpendicular to its velocity (tangential to the curve). Thus passengers (and other objects) will be subject to centrifugal force. Were the centrifugal force reduced or removed within the cabin objects would not stay within the cabin. None of this has anything to do with who or what is driving. It's just the laws of motion.

Yes my statement was incorrect, centrifugal force cannnot be reduced, however by changing the body roll, centripetal force exercised on the body can be shifted for a more comfortable ride

Advertisement

 

[ajdelange]

Now that I'll buy. But do note that I bobbled the math when I took the derivative. The post has been corrected,

 

[MikeThier]

From what I’ve seen of FSD beta, it will slow down for speed bumps. Perhaps with the linked suspension, it will adjust the ride instead of the speed over speed bumps.

Hmmm. I'm older and I like to control my own speed, but I guess that will be a user setting.

 

[Newton]

Says the guy who isn't getting fsd.

Haha, but I recently added fsd. Decided to get it if i can afford it. I didn't know about the supercomputer they will be using to help train it. Which I think will greatly decrease the time needed to get level5

 

[Billyboy]

Lowering the vehicle at highway speeds will give a bit better range.
If someone mentioned this, I didn’t see it.

 

[Sirfun]

This makes a lot of sense. At highway speeds it could lower the CT to the lowest setting with soft comfy ride. Then on a curvy mountain road, lower it and stiffen the suspension for the curves. Driving in normal stop and go traffic raise it, for better visibility and adjust suspension to deal with rough sections of road, pot holes, or bridges that have those rough transitions. How about adjusting the suspension based on how the truck is loaded or towing. Lot's of GOOD could come from this!

 

[Crissa]

In the Model S, they had it do that originally - go to its lowest on the highway. But that resulted in several instances where the car didn't make it over debris, including some fires.

I assume they're linking the feature to FSD so the computer can 'dodge' bad road surfaces and debris.

-Crissa

 

[Bill906]

My Jeep Grand Cherokee has “Quadralift” which is what they call their active air suspension. It has 5 levels that I have some control over, depending on the speed I’m traveling.

• Off Road 2 - Highest level. Will not go into or stay at this level if going above a certain speed. I think its 10 or 15MPH.
• Off Road 1 - 2nd highest level. I think the speed threshold is no faster than 25MPH
• Normal - Typical city driving.
• Aero- Lower than normal to increase aero dynamics. I cannot Select this level. Vehicle automatically goes to this level when above a speed. I think it’s 50MPH.
• PARK (aka Entry/Exit). Lowest setting. I have it set to automatically go to this level when car is put in park. Will not go to or stay at this level above a certain speed ( I think 20MPH).

Couple notes.:

• Vehicle will not lower if a door is open.
• It’s pretty slow. Roughly 10 seconds to go from one level to the next. Around 40 seconds to go from lowest level to highest level.

I, of course, do not know how the CT‘s active suspension will work, I can only spectuate it will be similar to the Jeep.

I was somewhat disappointed that it would not let me go to the highest level at highway speeds. If someone‘s traveling slowly in the left lane I thought a taller vehicle in their rearview mirror might help them realize their error.

 

[T3slaDad]

Haha, but I recently added fsd. Decided to get it if i can afford it. I didn't know about the supercomputer they will be using to help train it. Which I think will greatly decrease the time needed to get level5

Another user converted by the reality of FSD, I love it!!

It's really sad that the DOJO supercomputer isn't talked about more in streamline media. Tesla owns the world's fastest supercomputer (Tesla's is 1 petaflop, Japan is 2nd at 460 teraflops, less than half Tesla's), and its purpose is to single handedly train and master the FSD platform. I don't think Google or any other company is nearly that dedicated to FSD ?

Oh and did I mention, the DOJO isn't even fully active yet? The dojo sensei is currently stretching and warming up, and he will be ready to train his FSD grasshopper once the beta has rolled out to a majority of the fleet next year.

Just you wait, once the DOJO is fully online, it'll be a madhouse of progress that noone has ever imagined!!

 

[FutureBoy]

It's really sad that the DOJO supercomputer isn't talked about more in streamline media. Tesla owns the world's fastest supercomputer (Tesla's is 1 petaflop, Japan is 2nd at 460 teraflops, less than half Tesla's), and its purpose is to single handedly train and master the FSD platform.

Tesla owns the world's fastest supercomputer

That we know of.......

From what I understand, DOJO isn’t fully built out yet. And it wouldn’t surprise me at all if some TLA agency had larger hardware. Also, if Tesla were to go to a cloud approach, I’m sure that there are clouds out there (Google, Microsoft, Amazon) that have more compute power.

Not trying to diminish the hardware achievements of DOJO. It’s amazing and will most likely revolutionize FSD. I am very hopeful that it is going to be everything Tesla says it will plus a whole lot more.

 

[T3slaDad]

That we know of.......

From what I understand, DOJO isn’t fully built out yet. And it wouldn’t surprise me at all if some TLA agency had larger hardware. Also, if Tesla were to go to a cloud approach, I’m sure that there are clouds out there (Google, Microsoft, Amazon) that have more compute power.

Not trying to diminish the hardware achievements of DOJO. It’s amazing and will most likely revolutionize FSD. I am very hopeful that it is going to be everything Tesla says it will plus a whole lot more.

Of course, cloud computing can be extremely powerful! And yes, we don't know of the faster supercomputers that *may or may not* exist (Eagle Eye, anyone??). But regardless how fast other computers are, I bet you that the combined computing power toward any other company's efforts of FSD is bar none to Tesla's computing efforts.

Tesla has shown time and time again that they care the most about making fully-adaptable FSD a reality. If anyone wanted to put together a better solution for FSD computing, it can't just be done overnight. This stuff takes years to develop and produce, and Tesla is at the end of those years of ramping. Others, not so much. They either haven't pushed or have pushed, but not to the extremes of Tesla.

I love that multiple solutions are being developed, though! It's nice to see that everyone is thinking outside of the box, but now is the time for them to focus and go all-in on one technology instead of playing with lots of options still if they want to catch up to Tesla.

 

[Crissa]

We're at an inflection point where the investment needed for a third party to come into self driving that if they haven't been running, they'll never catch up.

Hopefully, this sort of program will be able to be more open platform in the future. So we can have competition and impartial inspections.

-Crissa

 

[FutureBoy]

We're at an inflection point where the investment needed for a third party to come into self driving that if they haven't been running, they'll never catch up.

Hopefully, this sort of program will be able to be more open platform in the future. So we can have competition and impartial inspections.

-Crissa

The thing about most AI is that the code doesn’t tell you much about what the behavior will be. In most implementations the only way to know is to actually run a test/simulation. There have been AI implementations that are auditable to the point of showing what the contributing factors were for each decision but even in that case the audit comes post decision. I’m not privy to what Tesla is doing on the AI front. From what I’ve seen most auditable AI systems were created to track down causes of things that were not understood like racial bias in facial recognition. Maybe Tesla has implemented some sort of auditing. Seems to me though that with a training system like DOJO they would be able to create some auditing in DOJO and then run simulations there instead of on an actual vehicle.

I’d think this would also route around most of the code review/auditing that someone or a government might want to do on a vehicle.

But if we are talking a review to find sleeper agent spy code, that would most likely not be in FSD itself as that would be more about telling the vehicle where to go in the way of directions like normal mapping directions. FSD is more about nanosecond by nanosecond determining what the next step is to fulfill the mandate it’s been given from the driving directions. Though, if the directions were to run over a specific individual, FSD would need an override to allow it to ignore object avoidance.

 

[Crissa]

Well, from their diagnostics, it does seem like they've implemented auditing.

It's very stupid not to have auditing in your AI. Otherwise, how are you supposed to give it the right training data?

-Crissa

 

[FutureBoy]

Well, from their diagnostics, it does seem like they've implemented auditing.

It's very stupid not to have auditing in your AI. Otherwise, how are you supposed to give it the right training data?

-Crissa

From what I understand of Tesla’s approach, the right training data essentially comes down to all of it.