• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

YACT: "self-learning" auto tranny's?

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Originally posted by: Heisenberg
Well how the hell do you think it monitors what the car is doing? Psychic powers? It uses input from the various sensors to determine what your "style" is.
Click to expand...

I'm aware that it requires sensors to know what's going on, however that's not adaptive.

If the program is just "if acceleration is X, shift at 2500RPM, if acceleration is Y, shift at 5000RPM", there is nothing adaptive about that. You can drive the car for a year and the program is still gonig to shift at 2500RPM with acceleration X and 5000RPM with acceleration Y.

For the car to actually be adaptive, so that it actually changes it's program to suit its driver, then the shift points need to change over time. So, if you by a car and drive it for a few thousand miles, then maybe at acceleration X it shifts at 3000 instead of the 2500 that it was set at from the factory. This would be different from the shift points in your friends car of the exact same model based on his driving style.

I see absolutely no reason for the car to do that. And certainly the fact that the car has accelerometers in the wheels doesn't mean that it does that.

Also, both neural networks and bayesian algorithms, which are the most common forms of computer learning, require feedback to verify that thier adjustments are valid. A transmission controlled by a neural network would need a target for a "good shift" that it was trying to get close to in order to actually learn, and I have no idea what the criteria for a "good shift" would be, especially without the driver providing any feedback to the transmission.
 
Originally posted by: LocutusX
Originally posted by: notfred
The buick transmissiont hat Heisenberg posted says it adapts to your driving style, but it doesn't actually say anything about how it works.
Click to expand...


I actually share your skepticism, notfred. That's why I posted this thread.

I'm trying to find out if this is a real feature that actually existed on the previous-generation Civic, and if so - how to notify the "engine computer" that there's a new driver on board and that it needs to forget how the previous old woman drove the car...

Someone said yank the battery cable for a bit... I guess that makes sense, unless there's a lithium battery connected to the car electronics for "backup purposes" 😛
Click to expand...


I would imagine it is not on a Civic. Caddys, and some lexus, M-B cars do.
 
Originally posted by: notfred
Originally posted by: Heisenberg
Well how the hell do you think it monitors what the car is doing? Psychic powers? It uses input from the various sensors to determine what your "style" is.
Click to expand...

I'm aware that it requires sensors to know what's going on, however that's not adaptive.

If the program is just "if acceleration is X, shift at 2500RPM, if acceleration is Y, shift at 5000RPM", there is nothing adaptive about that. You can drive the car for a year and the program is still gonig to shift at 2500RPM with acceleration X and 5000RPM with acceleration Y.

For the car to actually be adaptive, so that it actually changes it's program to suit its driver, then the shift points need to change over time. So, if you by a car and drive it for a few thousand miles, then maybe at acceleration X it shifts at 3000 instead of the 2500 that it was set at from the factory. This would be different from the shift points in your friends car of the exact same model based on his driving style.

I see absolutely no reason for the car to do that. And certainly the fact that the car has accelerometers in the wheels doesn't mean that it does that.

Also, both neural networks and bayesian algorithms, which are the most common forms of computer learning, require feedback to verify that thier adjustments are valid. A transmission controlled by a neural network would need a target for a "good shift" that it was trying to get close to in order to actually learn, and I have no idea what the criteria for a "good shift" would be, especially without the driver providing any feedback to the transmission.
Click to expand...
No, just using the sensor input to shift at a certain point is not adaptive. But the whole point is that the ECU monitors the sensor input and then presumabely using some kind of adaptive logic programming to decide when to shift. I would then guess it uses the sensor data as feedback to determine if that shift was "good". However, as I'm not familar with how the programming specifically works, I'm not going guess any more than that.
 
Originally posted by: Heisenberg
No, just using the sensor input to shift at a certain point is not adaptive. But the whole point is that the ECU monitors the sensor input and then presumabely using some kind of adaptive logic programming to decide when to shift. I would then guess it uses the sensor data as feedback to determine if that shift was "good". However, as I'm not familar with how the programming specifically works, I'm not going guess any more than that.
Click to expand...

I just can't see what the benefit to doing that would be. Assuming two different drivers were accelerating at the same rate and travelling at the same speed, I don't see any reason why it would be better for one car's transmission to shift at a different point that the other.
 
Originally posted by: LocutusX
Hey all,

While doing some car shopping on the weekend, I was trying out two different (used) 2002 Civic EX sedans both with auto. During the test drive I took them both down straightaways and applied a fair amount of pressure to the throttle, without flooring it. What I noticed was that the first one seemed to rev until ~5000rpm before upshifting, while the second one upshifted at ~4000rpm.

Why would there be a difference? Someone mentioned that auto tranny's from 2000+ cars are usually "self-learning" meaning they adapt to the style of the driver. If this is true, then how would I get the cars to "un-learn" if I bought it, since I would want the car to adapt to *my* style?
Click to expand...
They're not "self-learning." The issue is that modern ECU's and throttle position sensors are very sensitive to very small differences in throttle position in that range of acceleration (50% to 75% throttle). What really happened is that you weren't applying the exact same amount of throttle even though you thought you were. No fault to you though. Both were used cars with throttle cables, so you could have put your foot in the exact same place but indicated a slightly different throttle position to the TPS due to physical wear in the cable.
The only area where modern ECU's are adaptive is in spark timing.
 
Originally posted by: Vic
Originally posted by: LocutusX
Hey all,

While doing some car shopping on the weekend, I was trying out two different (used) 2002 Civic EX sedans both with auto. During the test drive I took them both down straightaways and applied a fair amount of pressure to the throttle, without flooring it. What I noticed was that the first one seemed to rev until ~5000rpm before upshifting, while the second one upshifted at ~4000rpm.

Why would there be a difference? Someone mentioned that auto tranny's from 2000+ cars are usually "self-learning" meaning they adapt to the style of the driver. If this is true, then how would I get the cars to "un-learn" if I bought it, since I would want the car to adapt to *my* style?
Click to expand...
They're not "self-learning." The issue is that modern ECU's and throttle position sensors are very sensitive to very small differences in throttle position in that range of acceleration (50% to 75% throttle). What really happened is that you weren't applying the exact same amount of throttle even though you thought you were. No fault to you though. Both were used cars with throttle cables, so you could have put your foot in the exact same place but indicated a slightly different throttle position to the TPS due to physical wear in the cable.
The only area where modern ECU's are adaptive is in spark timing.
Click to expand...

yea! Someone agrees with me 🙂
 
Originally posted by: notfred
Originally posted by: Heisenberg
No, just using the sensor input to shift at a certain point is not adaptive. But the whole point is that the ECU monitors the sensor input and then presumabely using some kind of adaptive logic programming to decide when to shift. I would then guess it uses the sensor data as feedback to determine if that shift was "good". However, as I'm not familar with how the programming specifically works, I'm not going guess any more than that.
Click to expand...

I just can't see what the benefit to doing that would be. Assuming two different drivers were accelerating at the same rate and travelling at the same speed, I don't see any reason why it would be better for one car's transmission to shift at a different point that the other.
Click to expand...
I dunno, but presumabely is has some benefit (fuel economy, or keeping the motor at the right rpm for the most torque perhaps) otherwise you wouldn't think the manufacturers would go to the expense of developing it.

Edit: I guess the other issue is exactly what the definition of "adaptive" or "self-learning" is.
 
Originally posted by: Heisenberg
Originally posted by: notfred
Originally posted by: Heisenberg
No, just using the sensor input to shift at a certain point is not adaptive. But the whole point is that the ECU monitors the sensor input and then presumabely using some kind of adaptive logic programming to decide when to shift. I would then guess it uses the sensor data as feedback to determine if that shift was "good". However, as I'm not familar with how the programming specifically works, I'm not going guess any more than that.
Click to expand...

I just can't see what the benefit to doing that would be. Assuming two different drivers were accelerating at the same rate and travelling at the same speed, I don't see any reason why it would be better for one car's transmission to shift at a different point that the other.
Click to expand...
I dunno, but presumabely is has some benefit (fuel economy, or keeping the motor at the right rpm for the most torque perhaps) otherwise you wouldn't think the manufacturers would go to the expense of developing it.
Click to expand...

How would fuel economy be better based on how a specific driver drives? Or for that matter, how would the torque curve of the engine change depending on the driver?
 
Vic, some throttles are fly by wire -- no cable at at all. So even those are only spark timing adjustments? I thought it was that it changed the shift point if it detected that you were an agressive driver?

Basically, they way the dealer explained it to me..my volvo has a few "programs" in its computer. If you drive like a nanny, it sets it to shift early to save gas. If you drive agressive for a day or two, it sets it to a program that shifts more agressively. It's not an AI, but it has different set points for different styles of driving.
 
From everything I've ever read, under full throttle acceleration any ECtransmission should shift at the manuf.'s safe redline regardless of the programming. The shiftpoints that are programmed in are for the partial throttle acceleration.

For example (older tech) on our 96 Camry there's a button you can press to change the programming (when enabled it'll shift under partial acceleration at, say, 5000 RPM vs 3000 RPM not enabled; in either mode, in WOT, the transmission shifts at maximum RPM).

On newer cars, they got rid of the button.

 
Originally posted by: LordSegan
Vic, some throttles are fly by wire -- no cable at at all. So even those are only spark timing adjustments? I thought it was that it changed the shift point if it detected that you were an agressive driver?

Basically, they way the dealer explained it to me..my volvo has a few "programs" in its computer. If you drive like a nanny, it sets it to shift early to save gas. If you drive agressive for a day or two, it sets it to a program that shifts more agressively. It's not an AI, but it has different set points for different styles of driving.
Click to expand...

The chance that your car dealer knows how the transmission programming works is a lot lower than the chance that your Best Buy salesman knows the difference between a P4 and a Celeron, and you already know how unlikely that is. 🙂
 
Originally posted by: LordSegan
Vic, some throttles are fly by wire -- no cable at at all. So even those are only spark timing adjustments? I thought it was that it changed the shift point if it detected that you were an agressive driver?

Basically, they way the dealer explained it to me..my volvo has a few "programs" in its computer. If you drive like a nanny, it sets it to shift early to save gas. If you drive agressive for a day or two, it sets it to a program that shifts more agressively. It's not an AI, but it has different set points for different styles of driving.
Click to expand...
A 2002 Civic EX auto is not "drive by wire."

Dealer advice should always be taken with a grain of salt. There are "sport modes" in some auto trannies, but it's not "self-learning" or even remotely AI.
 
It's not so much a 'feature', but a byproduct of how the whole system works. It doesn't have AI or even a program to do this. The ECu controls a bunch of different systems that talk to it constantly. Instead of reacting immediately to what you're doing, it keeps a record of what you've done in similar situations. Based on that record, it pre-applies a bunch of setting to things like the fuel pump, injectors, plug coils, etc. So if you regularly accelerate fast from a sop light, it knows that you're gonna mash the gsa shortly after the car is completely stopped and gears up for that presetting the timing, fuel/air mix, etc for a quick takeoff. If you slowly pull away, it adjusts the timing, fuel/air etc. for that. It'll even take record of how long you accelerate (say at highway speeds to pass someone) and move to that mode when you punch it.
Again, these aren't AI programs or special equipment, it's just a byproduct of pretty much any ECU system. And yes, disconnecting the battery for 30 min or so will reset the ECU and lose these stored algorithms.
 
Originally posted by: Heisenberg
Here's a few links to look through.
One
Two
Three
Four
Five

So either the manufacturers are lying, or the definitions of "adaptive" are different.
Click to expand...

Here's the sum total of all the information on "adaptive shifting" from all of those links:

"Some computerized transmissions even learn your driving style and constantly adapt to it so that every shift is timed precisely when you would need it."
Click to expand...

INVECS II is a computer-controlled automatic transmission which has the ability to "learn" your driving style...
Click to expand...

"Some computerised autos even have learning function (some uses Fuzzy logic). They memorise the driving habit of the driver through reading the input from throttle, rev counter etc. Then adapt their programs to suit the driving style of the driver."
Click to expand...

However, fuzzy logic is NOT adaptive and doesn't learn, so I'm not going to put a lot of faith in this source. The poor English certainly isn't helping, either.

So yes, some manufacturers are saying that their transmissions can learn, but there's no information about how it works or what it actually improves, and honestly it wouldn't surprise me if it's more marketing than anything else.
 
Originally posted by: notfred
However, fuzzy logic is NOT adaptive and doesn't learn, so I'm not going to put a lot of faith in this source. The poor English certainly isn't helping, either.

So yes, some manufacturers are saying that their transmissions can learn, but there's no information about how it works or what it actually improves, and honestly it wouldn't surprise me if it's more marketing than anything else.
Click to expand...
That's just what I found in 5 minutes. I'm sure if you looked harder you could find more, although I doubt any company will publish exactly how their system works for business reasons. If you want to disagree that it exists for whatever reason, that's your prerogative.
 
Originally posted by: notfred

However, fuzzy logic is NOT adaptive and doesn't learn, so I'm not going to put a lot of faith in this source. The poor English certainly isn't helping, either.

So yes, some manufacturers are saying that their transmissions can learn, but there's no information about how it works or what it actually improves, and honestly it wouldn't surprise me if it's more marketing than anything else.
Click to expand...

Well then let my personal experience speak, over three months of driving a brand new mercury sable (2003) I have experienced the transmission learning the drivers habits of acceleration and adjusting accordingly to provide a much smoother experience. There are adaptive transmissions out there, its a simple thing and nothing amazing and complex and foreign as some people here believe.

Dont believe me? I dont care. Fact is they exist and work.

 
Originally posted by: Heisenberg
Originally posted by: notfred
However, fuzzy logic is NOT adaptive and doesn't learn, so I'm not going to put a lot of faith in this source. The poor English certainly isn't helping, either.

So yes, some manufacturers are saying that their transmissions can learn, but there's no information about how it works or what it actually improves, and honestly it wouldn't surprise me if it's more marketing than anything else.
Click to expand...
That's just what I found in 5 minutes. I'm sure if you looked harder you could find more, although I doubt any company will publish exactly how their system works for business reasons. If you want to disagree that it exists for whatever reason, that's your prerogative.
Click to expand...

I'm just saying that I wouldn't be surprised if the auto industry's definition of "adaptive" isn't quite the same as what other people's idea of "adaptive" is. Kind of like how the hard drive industry's idea of a gigabyte is about 7.3% smaller than everyone else's definition of a gigabyte.
 
Originally posted by: Metron
So notfred, vic and amdskip, do you think the earth is flat?

Sun orbits the earth?

No, it's just can't be! :shocked:
Click to expand...

HAHAHA!!! This is quite possibly the worst use of strawman I have ever seen on ATOT. That's saying a lot BTW.

"Adaptive" automatic transmissions is basically a marketing phrase. You might want to stop and think about what the basic purpose and function of an automatic transmission is. If it wasn't to "adapt" to changing condition like throttle position, intake manifold vaccum, rpm, etc., then they wouldn't work very well, would they?

You basically have no idea WTF you're talking about so you're resorted to strawman and google search. Pretty sad.
 
Originally posted by: Aquila76
It's not so much a 'feature', but a byproduct of how the whole system works. It doesn't have AI or even a program to do this. The ECu controls a bunch of different systems that talk to it constantly. Instead of reacting immediately to what you're doing, it keeps a record of what you've done in similar situations. Based on that record, it pre-applies a bunch of setting to things like the fuel pump, injectors, plug coils, etc. So if you regularly accelerate fast from a sop light, it knows that you're gonna mash the gsa shortly after the car is completely stopped and gears up for that presetting the timing, fuel/air mix, etc for a quick takeoff. If you slowly pull away, it adjusts the timing, fuel/air etc. for that. It'll even take record of how long you accelerate (say at highway speeds to pass someone) and move to that mode when you punch it.
Again, these aren't AI programs or special equipment, it's just a byproduct of pretty much any ECU system. And yes, disconnecting the battery for 30 min or so will reset the ECU and lose these stored algorithms.
Click to expand...
This is basically correct, except it's not even that complicated. Essentially, based on previous conditions, the ECU stores a single hexidecimal number in its memory to remind it of its overall state, along with a recorded matrix of what it expects for a given throttle position, rpm, and timing.
ECU logic is extremely limited -- at the most basic core of logic -- "given these circumstances, this happened in the past, therefore given the same circumstances, the same thing will happen again, therefore I should act accordingly."

This is your "adaptive." No more.


edit: It seems to me that the mistake people are making here is that they are assuming that the OP's test was somehow scientific. As I already pointed out, it was not, because modern ECU's are very sensitive to throttle position in that mid-range, and there is no way of knowing that he was at the exact throttle position for both cars in both runs (unless he had a laptop plugged in with his test-and-tune program running, which he didn't mention, so I'm assuming he did not).
FYI: proper rolling testing of a car's operation involves rolling onto the throttle -- light at low rpm, ~60% at mid-range, and then full throttle just before 4000. This tests the engine under all its most difficult tuning conditions in one run.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top