Lag in Cybertruck's drive-by-wire system
twitter.com
twitter.com
The first parts of the movement are actually quite responsive but don't result in a huge change.
If the wheels moved as the drive dictated the car would flip over. You can't just magically make all the forward momentum go lateral without relying on road and tire friction to get the job done. This can hit limits / you just end up skidding the tires and going ass over end instead of turning because you yanked the wheels too dramatically.
It would be like riding a bicycle and then turning the handlebars completely to one side while going at a good pace forward.
You could argue why isn't the car just doing it because they're at a standstill and it should be safe to do so? It's probably a sane default in case there are no sensors / no velocity reading / etc.
By the way, if this is true, it is beyond stupid...
Do you mean “… at high speed?” To me, it looks like the person just wants this car to steer the way literally every other car does. Your bicycle analogy — yes, that’s how they work, so pretty quickly you learn how to do that.
The demo shows the person stearing couldn't apply urgency in directing the wheel, sounds dangerous. Onus should be on the driver being capable of making the steering angle appropriate for the conditions, defending eventually consistent steering is comical, now drivers must account for where and by what speed the wheel angle reaches the desired position?
That's how things work though. Momentum and material strain mean that nothing reacts instantly, and that changes with your speed, steering angle, and rate of change in speed and steering angle. So yes, all drivers must account for the where and by what speed the wheel angle reaches the desired position.
I'm all for evidence showing the latency shown on the video is comparable to competing implementations, but I'm not interested in lazy arguments that all momentum is eventually consistent as a defence.
It's a vehicle you can buy right now, and this clip comes from a review (thanks rainbowzootsuit).
So the argument is that even with critical sensors offline or not providing meaningful data, the automation should continue to operate as normal? I believe Boeing has said much the same. The outcomes they've had by it lead me to doubt the merit of the argument.
If I try to turn the front wheel of my bike too far left or right it does not let me. Further, this is accomplished by a rubber bumper that continues to push for a short time after I let up, but it would be absurd to call this "lag." Normal operation of the bicycle never gets close to those limits.
The question is whether this video is showing typical system lag -- in which case, yeah, I agree, it's a problem -- or whether it is showing limiting behavior in which case getting upset about it makes as much sense as getting upset over the rubber bumpers on my bike.
Not sure what you're referring to here - I've got multiple bicycles and the headset bearing allows them to spin around 360 degrees, the only thing preventing them from going further is the cables and hydraulic hoses.
>Further, this is accomplished by a rubber bumper that continues to push for a short time after I let up
What bumper do you mean? Your handlebars, stem, and fork should be moving as a solid unit, with minimal deformation under any reasonable conditions. There's very little elasticity in the system, perhaps the biggest source is just the spokes (or the grips if they're quite thick). The behavior exhibited in the video is way more than any properly operating bicycle.
Your rims might, your tires don't. Check out about halfway down this page:
If they don't something is broken.
>your tires don't.
I did say "wheels", which generally speaking means the rim. Yes, I'm well aware of tire deformation and contact patch mechanics, but that's not the issue here - the rims on the truck are not moving in sync with the wheel. It's just not acceptable.
The speed of sound in steel and roughly 1m of steering column put the theoretical minimum latency at 0.2ms. Manufacturing tolerances and material strain put minimum bounds on accuracy as well. Your entire suspension, down to the rigidity of the individual linkages, influences the steering of your vehicle.
> Yes, I'm well aware of tire deformation and contact patch mechanics, but that's not the issue here...
I think it's very relevant to know that cars already take the direction of the tires/rims/wheel as more of a suggestion. The "tightness" of the steering and the suspension varies drastically between vehicles. So what is the issue here? How it drives. I don't have one, but a common comment I've found in reviews is that it's "too sensitive." I think this is an imagined problem.
Perhaps saying that the wheels move "exactly" as the driver intends in a rack and pinion system is too absolute, but if you can't see or feel this delay then I'd argue that to the human driving it's irrelevant. You can see the lag in the video, which admittedly is slowed down, but if we assume the original frame rate was 60fps then there's a minimum of a 16.7ms lag and I'd estimate it's at least 3-4 frames so more like 50.0-66.7ms, which is absolutely noticeable. Why should we accept performance which is clearly worse than a mechanical system? This is a delay that exists on top of all the other factors you've mentioned, not in lieu of them. It casts doubts on the functional safety of the system.
I could push your argument forward a bit more: the same vehicle can have tightness change as speeds goes up: electric power assistance progressively reduces as speed goes up. Why? because electric power steering is so powerful that it would be too easy to quickly turn wheels at speed. So the helper reduces, making it harder to turn the wheel in the same way that it became harder before power steering because physics.
I needs more force, but the lag doesn't change, and neither is the gear reduction ratio, nor the absolute angle of the steering wheel to the actual angle of the wheels relative to the body.
> I think it's very relevant to know that cars already take the direction of the tires/rims/wheel as more of a suggestion.
I see this vehicle as introducing a fundamental breaking change in how automobiles are being operated:
a) previously: you the driver physically control the car parts. You drive, operating the car by transforming high level intent to low level commands. You're the pilot sitting at the helm and your job is made easier with some helpful systems (ABS, ESP, cruise control, automated transmission, lane assist, what have you) that you can shunt.
b) this vehicle: you communicate intention "I want to go in this general direction" / "Set a course to Rigel VII. Maximum warp" and someone else follows your instruction and drives the car for you. The physics is abstracted away. You're not the pilot, you're the captain, and captains don't go to helm.
Up til now we've seen a) being progressively automated, working up towards full automation.
This vehicle instead appears to be designed to be designed downwards from full automation, offering a degraded mode of operation from full automation that looks like a) but is fundamentally different from a).
You can't shunt systems because that would prevent the actual pilot from operating the vehicle. In fact the pilot could very well do things that would be surprising and you would not be able to override it because your role is higher up.
I fully expect this design to lead to the cap'n turning the wheel slightly left but the pilot "refuses" to comply and slaloms around a bunch of pedestrians.
In any case, this vehicle has rims and tires; what's being discussed here only adds lag to the existing system, which appears to be several orders of magnitude higher than the above physical process.
But what's apparently worse here is that there's no link (whether physical or digital): the video shows the driver has moved the steering device all the way to lock yet the wheels are only halfway through and then continue moving towards lock in spite of the steering device being completely still.
> If the wheels moved as the drive dictated the car would flip over.
Non drive by wire cars react instantly by virtue of having wheels and steering physically linked. One can tip over such a car in the same way as you describe, the only thing preventing it is a) physics which makes it a bit harder to apply enough force and b) physics which gives the driver feedback that they're doing it wrong.
Which leads us to the crux of the issue here: this DBW implementation decouples the steering device not just physically in action but in dynamics and feedback as well: It's literally a joystick masqueraded as a steering wheel, with apparently little to no force feedback, and very obviously force feedback completely decorrelated from the actual angle of the wheels.
Taking a side road through the bicycle example: one doesn't turn a bicycle by turning the handlebar, one turns a bicycle by angling the whole apparatus, and the front end follows due to its geometry.
The same principle applies to a car's front suspension: the geometry of it makes it resist being turned, which provides quintessential feedback to safe operation. Without this, a critically important signal is removed. A second one is removed as there is jo way to know how angled wheels are relative to the car's body. Throw in latency into the mix and you get a surefire way to create an overcorrection feedback loop and be completely unable to react in even slightly non-ideal situations.
This means that for any sort of safety that should be by design of the chassis and suspension the vehicle relies entirely on ESP, when ESP should be a last resort fallback when the normal mode of operation fails. Instead there is no fallback nor redundancy and ESP MUST work or this vehicle becomes a death machine at the slightest sideway skid that would be perfectly manageable otherwise, even without ESP.
> sane default
This means the vehicle changes its dynamic reaction to driver input. Training aircraft pilots about normal law vs alt law is already quite something, I don't think training drivers about similar things is going to be very successful.
That's a bit morbid, but if you can hack the computer or manipulate sensor input to change the car comportement, it'll be easier, more effective and more stealthy for assassination.
Driving Tesla Cybertruck - Overhyped or the Pickup Reinvented?
https://www.youtube.com/watch?v=TeqE1kOO3dE
From about 10m into the video.
Needless to say, if you put someone at the wheel of a vehicle with regular power assist steering, the wheels aren't going to move any faster.
[1] Or others. Tesla is not the first company to offer steer-by-wire, just the first to engender this kind of nonsense debate about every decision they make.
Our 2005 DARPA Grand Challenge vehicle, a Polaris Ranger ATV, looked like that in manual mode. The steering wheel was drive by wire, and used a simple PI controller. Stationary, with the steering system fighting static friction, the normal P value useful for driving is too low, so there's lag. The I term cranks up the force until the error goes to 0, but that takes time. Our remote dashboard had two pointers on the steering meter - commanded steering angle and actual steering angle. We could watch the lag. Half a second of lag was OK, because we didn't do any aggressive maneuvers. Manual mode was just for positioning the vehicle before going autonomous.
That slowdown at the end of the Tesla video looks like normal operation of a PI controller.
Does Tesla have steering force feedback? If the steering wheel is leading the wheels, the steering wheel should be pushing back on the driver if there's significant error.
I like cars, I have built and raced cars. I have owned sports cars, boring hybrids, and a Tesla. I have tuned ECUs and adjust BCMs… cars do not behave as linearly and directly as a lot of Internet engineers seem to think they do.
If you watch closely in this video, the steering wheel turning and the wheel starting to turn happen almost instantaneously, there is no initial input lag. The lag is between the steering wheel and wheel reaching center, and that’s dictated by an intentional acceleration curve. This is what I would have expected in a steer-by-wire system.
There are things worth complaining about with steer-by-wire and I don’t think this is one of them.
Cool car background by the way! I just wonder why using steer-by-wire, what's wrong with modern power steering?
There could be a number of reasons for this but no modern car has a direct connection between the pedal and the throttle bodies, so it isn’t inherent to drive-by-wire. It can probably be tuned out of the S3 in software pretty easily. I haven’t driven one so I won’t claim there is no lag in the Cybertruck’s steering, but I find it pretty unlikely there is observable lag in the initial response compared to other vehicles in its weight class, which aren’t particularly responsive.
The three reasons I think are most relevant to Tesla:
1. Packaging - removing the steering column is a pretty dramatic reduction in the number of parts in the front of the car and makes it possible to reuse that space. It may make it easier to design the front crumple zone as well.
2. Extreme variable steering ratios - you can tune the steering ratios much further with a fully by-wire system. In a large vehicle it makes it easier to maneuver at low and high speeds with less effort for the driver.
3. If autopilot were ever to come to fruition, the steering wheel won’t have to move while autopilot turns the car.
Regarding the S3, and Audi's in general, I know that I am one of the few people I know who is really annoyed by that.
So my guess is this tweet with one short video and no additional evidence or reasoning behind their claim is wrong.
this topic has been beaten to death throughout all aspects of computing, it always goes like this:
1. someone complains about input lag, because he feels it and it's annoying and prohibitive
2. other people who are slow and / or have misconfigured systems with such worse problems that they don't notice this disagree
3. 7 years later its proven to a wider audience that the lag is there and all the people who said it was there were right
4. the company sells an all new "low input lag" model (which doesn't actually require any money to implement, aside from a few hours dev time)
I get that it's "safer" for your typical driving conditions, but I'd imagine trying to get out of the way in a hurry it could be a problem... Knowing that's the case now, I kind of expect it, so it's changed my driving habits slightly and has made me plan ahead more than normal.
But I can't imagine how actual input latency would feel after that experience.
Just no road feedback...
Even if this wasn't by design and is a limit of the "power" of the power steering system unable to "keep up", so what? Try going 0.5mph and see how it feels.
Also notice its not LAG. It starts turning THE MOMENT you turn the steering wheel, it's rate of change just isn't the same. That's not latency. The vehicle still turns the wheels when you ask it to, even when stopped. In what scenario do you need a stopped vehicles tires to steer from lock to lock in under 3 seconds?
There is plenty to complain about the cybertruck. This aint it, chief.
Try it yourself here to see how bad that feels: https://www.skytopia.com/stuff/lag.html
That page also links to a great video on what it's like on a touchscreen: https://www.youtube.com/watch?v=vOvQCPLkPt4
I really wouldn't want highway drivers in that monster truck to have any barriers to their reactions, especially with its nontraditional crumple zone :(
There isn't really any excuse for this.
It's hardly a performance car so perhaps it's not a big issue, but it really should not the there and should be fixed.
There's not actually THAT much latency (delay between input and response). There IS some acceleration curve being applied (X degrees of input = X^Y degrees of steering, buffered over time), but that's totally different. It's probably safer this way vs suddenly jerking the actual wheels back and forth.
The tires start moving immediately, so it looks more like a limit in how quickly they change direction. It appears laggy because they don’t finish the movement when he does. They do respond right away when he shifts his arms a little in the middle of the video.
I’m only half joking. I’d be worried in the EU because they don’t drive around mostly in SUVs like we do. A cybertruck versus a panda would not go well.
And that's for all cars operating on public roads, not just registered, in an area. It might be legal to drive a cybertruck where one lives, but not in a neighboring city or state. Moving fast and ignoring regulations on Tesla's part only works for their customers until regulatory agencies encounter a reason to pay attention.
Every car for the last 150 years has had mechanical connection from the steering wheel to the tires and nobody is talking about the reliablity and safety concerns. The absolute state of hacker news.
Same sort of thing with people that complain about iterm being too laggy.
There's standards for control like MIL-STD-1472 which I'll bet this meets.
I'm sure none of them are "the highly redundant sensor fusion system with absurd levels of automation and sensing, the operator should not be able to notice any lag"
unless we're playing soccer / free-kicks, in which case, yeah, overengineer the hell out of it, sport.
ps, chapter 5 precisely specifies tolerances and minimums for things like latency of specific controls, for a control system for heavy machinery intended to be controlled by a human... and thanks for your input.