Investor Engineering Discussions

[Gigapress]

It's only speculation at this point and ignores the added engineering required for what is basically a party trick. I kind of hope they don't waste the time enabling this.

In the video, a clip of the Cybertruck prototype demonstrated four wheel steering at the Cyber Rodeo, so it looks like Tesla already did it or at least tried it on the prototype.

Also I think it’d be more than a party trick because pickup trucks of this size are pretty annoying to park, especially when taken into urban areas, and this will make parking a lot easier.

 

[mongo]

In the video, a clip of the Cybertruck prototype demonstrated four wheel steering at the Cyber Rodeo, so it looks like Tesla already did it or at least tried it on the prototype.

Also I think it’d be more than a party trick because pickup trucks of this size are pretty annoying to park, especially when taken into urban areas, and this will make parking a lot easier.

There is rear steering with a standard Ackerman setup that Tesla has a patent for as regards a electric rear steering box. Then there is a fully independent 4 corner setup that may be a bit much.
(Just going from video transcription)

 

[JRP3]

In the video, a clip of the Cybertruck prototype demonstrated four wheel steering at the Cyber Rodeo, so it looks like Tesla already did it or at least tried it on the prototype.

Also I think it’d be more than a party trick because pickup trucks of this size are pretty annoying to park, especially when taken into urban areas, and this will make parking a lot easier.

I was referring to the "Diamond steering" setup which independently steers each wheel. Steering the rear wheels together is all we've seen demonstrated as far as I know.

 

[Buckminster]

https://twitter.com/x/status/1561073796538400768

 

[Gigapress]

Their understanding of antilock brake systems seems lacking. Brembo system is an interesting idea, but not failsafe in the pure electic form. Though neither is hydraulic depending on the mechanical fault. Pedal pulsation could be eliminated from standard ABS setup.
Steering wheel already has absolute position encoders. Pure steer by wire is not failsafe, variable rate steering can be.
Tesla has a patent for an electric rear steering system (both wheels together). 4 wheel fully independent seems like a lot of extra parts for little gain.

How come steering and braking by wire isn’t fail safe? If electric control is good enough for fly-by-wire commercial aircraft like the 787 I don’t see why similar redundancy couldn’t be designed for a car.

After your and JRP’s points I agree that the independent wheel steering is probably overkill. Also I think Connecting the Dots missed that a tank turn can still be done in a non-diamond fashion as long as the rear wheels can turn at all. The front pair and rear pair would just need to be turned the same direction and rotated in opposite directions. Diamond not necessary.

 

[mongo]

How come steering and braking by wire isn’t fail safe? If electric control is good enough for fly-by-wire commercial aircraft like the 787 I don’t see why similar redundancy couldn’t be designed for a car.

After your and JRP’s points I agree that the independent wheel steering is probably overkill. Also I think Connecting the Dots missed that a tank turn can still be done in a non-diamond fashion as long as the rear wheels can turn at all. The front pair and rear pair would just need to be turned the same direction and rotated in opposite directions. Diamond not necessary.

I may expecting too much from fail-safe and drifting into the realm of fail-controlable.

So first acknowledging that the systems have many mechanical single points of failure. Both systems have to reduce the occurence as much as possible.

X-by wire assumes input sensors, communication link, processing, output actuators, and output position sensors are all operating and powered. For a Tesla, loss of pack (with DC output) and 12V aux battery will disable all electronic systems. It is single point of failure tolerant, if everything else has redundancy.

A car with mechanical steering and hydraulic brakes is controllable even in a vehicle with no power at all. That is, worst case failure mode not involving the base mechanical system.

Of course, autonomy is all X-by wire and has the levels of redundancy and fault detection needed to handle failures in as safe a way as possible which is going to mean removing all velocity in as safe a location as possible.

For planes, this approach is not acceptable as falling out of the sky is a bad thing. So they have double or even triple redundancy (which is still not always enough).

For both planes and autonomous fleets, it also means preventing operation if redundancy has been lost. This is viewed less favorably by an individual owner who may be willing to accept the loss of the self driving features if it means they can still get from point A to B.

 

[MC3OZ]

A car with mechanical steering and hydraulic brakes is controllable even in a vehicle with no power at all. That is, worst case failure mode not involving the base mechanical system.

One question would be :- How well and electric system could compensate and prevent a sudden unexpected failure?

Perhaps 16V battery or an oversized variant could provide an alternative power source.

Perhaps each wheel can have a capacitor to drive the electronic function, which can be charged from with either battery.

IMO all options are worth exploring, because ditching hydraulics does seem like it would make cars cheaper and easier to build.

 

[MC3OZ]

Even if the "Diamond Steering" video is likely to be well of target on some points, I do agree with the general premise.

Simply Tesla would be looking at all options to reduce cost and complexity when building the Cybertruck.

"This is the way we have always done it", will not be taken seriously

"This is the cheapest and best solution" will, that doesn't need to be a solution currently in use by most of the industry.

The aim isn't to achieve "party tricks", merely to demonstrate "party tricks" when good engineering means they are an added advantage of decisions already made.

Even as a party trick the tank turn is of limited value, any driver who genuinely needs to use a tank turn to get out of a situation can't drive a car in reverse.

 

[mongo]

Coming from a position of discussion, not argument.

One question would be :- How well and electric system could compensate and prevent a sudden unexpected failure?

Perhaps 16V battery or an oversized variant could provide an alternative power source.

Perhaps each wheel can have a capacitor to drive the electronic function, which can be charged from with either battery.

IMO all options are worth exploring, because ditching hydraulics does seem like it would make cars cheaper and easier to build.

Airbag modules, black box data recorders, and (I think) some electronically latched doors do this now. However, on loss of power to one brake, when is the correct action to brake that wheel?

Other than the centralized ABS module, hydraulic brakes are hella simple to make and produce. The outboard hardware is also quite durable.
Installation (other than bleeding) is fairly equivalent to harness routing. Replacement later can be a pain though.

Even if the "Diamond Steering" video is likely to be well of target on some points, I do agree with the general premise.

Simply Tesla would be looking at all options to reduce cost and complexity when building the Cybertruck.

"This is the way we have always done it", will not be taken seriously

"This is the cheapest and best solution" will, that doesn't need to be a solution currently in use by most of the industry.

The aim isn't to achieve "party tricks", merely to demonstrate "party tricks" when good engineering means they are an added advantage of decisions already made.

Even as a party trick the tank turn is of limited value, any driver who genuinely needs to use a tank turn to get out of a situation can't drive a car in reverse.

Best part is no part.
Cybertruck with quad motors already has per wheel braking ability via regen.

In duplicating modules, there is also the reliability probability problem. The odds of all modules working decreases as the number of modules increases, even as the odds of at least one module still working increase.

If chance of failure is f, and number of modules is n, then odds of a failure are (1-(1-f)^n).
Pure for illustration: 20% chance of failure per module
Chance of at least one module failure:
n=1: 20%
n=2: 36%
n=4: 59%
This is pary of why new large planes have two engines instead of 4.

 

[JRP3]

Perhaps each wheel can have a capacitor to drive the electronic function, which can be charged from with either battery.

Doesn't help if the signal wire is lost.

Simply Tesla would be looking at all options to reduce cost and complexity when building the Cybertruck.

Diamond steering adds complexity. There is no good reason to add it and plenty of reasons not to.

 

[MC3OZ]

Diamond steering adds complexity. There is no good reason to add it and plenty of reasons not to.

I wasn't suggesting Diamond steering, I think that is very unlikely.

I still think replacing hydraulics with electricity is worth considering, not essential, but an alternative.

If an electrical connection can be broken so too can a hydraulic connection, redundancy is more achievable via an electric/data coms path.

But there is no point in doing something new just for the sake of it.

If that new way of doing things makes a better product and/or saves money it is worth doing, even if initially there are some teething problems.

Any change to the vehicles steering/brakes will need to pass regulatory processes, which will not be easy.

But we are heading for a Robotaxi future, where feedback to the human driver via the wheel or pedal is less valuable, because there might not be a human driver.

 

[JRP3]

Do improvements in FSD ever get ported to standard autopilot, and have people seen autopilot improve over time? I find autopilot mostly useless because it hugs the center line too closely, especially on right hand curves, and it generally takes all curves too wide for my comfort. This is on clearly marked state routes. It's a bit better on multi lane interstates but only if traffic is minimal. If there were at least a way to bias it about 6 inches to the right it would feel less dangerous.

 

[navguy12]

Do improvements in FSD ever get ported to standard autopilot, and have people seen autopilot improve over time? I find autopilot mostly useless because it hugs the center line too closely, especially on right hand curves, and it generally takes all curves too wide for my comfort. This is on clearly marked state routes. It's a bit better on multi lane interstates but only if traffic is minimal. If there were at least a way to bias it about 6 inches to the right it would feel less dangerous.

And my issue on two lane highways also includes the “phantom braking“ whenever a large vehicle (or car towing a trailer) heading towards me in the oncoming lane when the road immediately ahead of me sweeps to the right…

 

[bkp_duke]

Do improvements in FSD ever get ported to standard autopilot, and have people seen autopilot improve over time? I find autopilot mostly useless because it hugs the center line too closely, especially on right hand curves, and it generally takes all curves too wide for my comfort. This is on clearly marked state routes. It's a bit better on multi lane interstates but only if traffic is minimal. If there were at least a way to bias it about 6 inches to the right it would feel less dangerous.

In a word: no.

Eventually: yes.

Autopilot and "drive on city streets" FSD are completely different software stacks.

That is what all the discussion is about merging things into "one stack". That merge is huge undertaking. It will get done, but when it happens it may actually make some things worse for a while.

 

[JRP3]

Autopilot and "drive on city streets" FSD are completely different software stacks.

Right I was just wondering if things learned from FSD are ever applied to autopilot. I guess autopilot is just being neglected until the FSD software stack can replace it completely.

 

[bkp_duke]

Right I was just wondering if things learned from FSD are ever applied to autopilot. I guess autopilot is just being neglected until the FSD software stack can replace it completely.

No, they are not. The point is the full stack, where at that time all of FSD's learning will be applied to Autopilot. Until then, only "critical" fixes are applied to Autopilot, and I haven't seen one of those in months.

 

[mongo]

Y'all are typing faster than me, so this is somewhat redundant.

To expand, there are three-ish stacks running around:
Navigate on Autopilot (NoA): works fairly well so they are not doing major changes at this point
Autosteer: the public release of FSD, includes traffic control recognition but not lane changes
FSD Beta: Pre wide release of Autosteer on City Streets, where all the code changes are happening

Once FSD Beta is mature (enough) it will replace Autosteer/ TACC. After that is further proven, it will replace NoA.

Key driver in the transtions is to have a system better than the previous one, i.e. no regressions.

The major changes in internal architecture prevent back porting improvements to the older code base. Fixes to labeling errors would help, but that infrastructure has also evolved (raw photon count, unified vector space).

 

[bkp_duke]

Y'all are typing faster than me, so this is somewhat redundant.

To expand, there are three-ish stacks running around:
Navigate on Autopilot (NoA): works fairly well so they are not doing major changes at this point
Autosteer: the public release of FSD, includes traffic control recognition but not lane changes
FSD Beta: Pre wide release of Autosteer on City Streets, where all the code changes are happening

Once FSD Beta is mature (enough) it will replace Autosteer/ TACC. After that is further proven, it will replace NoA.

Key driver in the transtions is to have a system better than the previous one, i.e. no regressions.

The major changes in internal architecture prevent back porting improvements to the older code base. Fixes to labeling errors would help, but that infrastructure has also evolved (raw photon count, unified vector space).

Also, of note, expect that only HW3 and higher cars will get the single stack. I don't see it as possible to put that stack on HW 2.5 or older cars because of the limitations of the NVidia hardware (and incompatibility with Tesla's neural net).

 

[mongo]

Also, of note, expect that only HW3 and higher cars will get the single stack. I don't see it as possible to put that stack on HW 2.5 or older cars because of the limitations of the NVidia hardware (and incompatibility with Tesla's neural net).

Oh, for sure! HW 3 upgrade roll out will go into full swing with the public Beta release.

Also, it seems that there may be both Beta and non-Beta versions of NoA, but they are very similar.

 

[willow_hiller]

Sandy's first review of the fully disassembled 4680 pack:

Long story short: lots of part reductions, and "poka yoke," or assembly-mistake-avoidance design techniques.