Computer shifted manual transmission

[PandaBear]

Quote:
Where computers struggle a lot is in mimicking smooth clutch engagement and gear shifting. Ungodly amounts of programming code has to be written to mimic this very simple, to humans, task that most every manual driver is capable of doing with a bit of practice. Slow moving traffic, stop and go and slippery conditions also present a challenge to the computers.


Human is very good at sensing and adapting to a feedback. Computer can do it too, but would require a lot of sensors, including audio (to interpret human's observing of the exhaust note and engine sound), pressure (to mimic what the left foot feel on the clutch), acceleration (to mimic the smoothness of shift that dictate your car's jerkiness), speed (to observe the progress of the shift). And these sensors have to be fast, cheap, and have wide enough of a range to work in all condition.

And of course, lots and lots of engineers' time.

Finding out which gear to stay in is easy, actually doing the driving, not so.

Originally Posted By: KrisZ
Originally Posted By: Vikas
so how does current slushbox solve all these problems? and is the fluid coupling aka torque converter and ant farm of narrow prone to clog passages and plethora of clutches constitutes the black magic which overcomes those obstacles?

I can buy that fluid coupling does indeed allow amount of slip which is far greater than that could be allowed using frictional standard clutch but in terms of anticipation or being in the right gear aspect should be identical in both the cases.


Fluid coupling plus clutch packs are activated by fluid pressure which is proportional to RPM, both of which are analog inputs/outputs. All slippage needs are taken by the fluid and torque converter, no computer algorithms are involved.
As far as holding gears or staying in one gear for no particular reason, both traditional and DCT type transmissions are at the same disadvantage and that is why most have a manual mode.


Fluid in torque converter acts as a cushion to soften the shift feel.

 

[KrisZ]

Originally Posted By: PandaBear
Fluid in torque converter acts as a cushion to soften the shift feel.


Fluid in the torque converter is only half of the story for soft/firm shifts. The pressure for the clutch packs is also regulated to provide smooth shifting at low speeds, low RPM's and firmer, quicker shifts at large throttle opening and high RPM's. There are lots of transmission tunes that can adjust shift firmness. In the old days you could do it by changing some check valves, but it was a permanent change.
Nowadays the pressure is regulated on the fly to provide optimal shift performance for most driving conditions, so some computing power and programming is used, but it's based on simple inputs and only has to control fluid pressure.

 

[A_Harman]

After having just driven my Camaro to work through a nasty snowstorm, it boggles my mind to think of the programming that would be necessary to make a computer mimic all the clutch modulation and careful gear selection that I had to do to keep the front end ahead of the back end.

 

[dishdude]

Smart cars have true automated manuals. When the car shifts, you get the exact same feeling that a manual gives - disengage clutch, shift, reengage clutch. It's weird!

 

[Vikas]

Panda and Kris has explained the difficulty for computer very nicely; thank you! Funny thing is I taught my wife and son how to drive manual transmission and both of them drive well without any clutch destroying habits. Neither are mechanically inclined but they got the hang of manual transmission fine. I understand that instincts are extremely difficult to teach to computer and indeed it would be tricky for computer to do city stop and go driving using mechanic clutch and manual gearbox but today computer is able to parallel park which none of us (me, my wife) can do without lot of external help! It does seem to me that degree of difficulty in parallel parking is lot higher than learning to drive manual.

 

[meep]

DOES ANYBODY REMEMBER '70's THE SUPERBEETLE?

the "autostick" had a large, 6" vacuum diaphragm directly connected to the clutch lever. It was a 3-4 (?) speed manual transmission with both a standard clutch plate and a fluid TC. there were only two pedals...gas and brake. it had a relatively large bell housing to hold it all in, and a stick shift on the floor.

I owned one but never drove it.... bought at a salvage yard for the engine. I believe the TC was good for idle and launch, and the clutch would disengage as you changed gears.

 

[KrisZ]

Originally Posted By: Vikas
I understand that instincts are extremely difficult to teach to computer and indeed it would be tricky for computer to do city stop and go driving using mechanic clutch and manual gearbox but today computer is able to parallel park which none of us (me, my wife) can do without lot of external help! It does seem to me that degree of difficulty in parallel parking is lot higher than learning to drive manual.


Actually self-parallel park is much easier if you break it down into inputs needed for a proper parallel park job. If you watch this video of Ford Focus parallel parking you will notice just how little input the computer needs to park the car.

There is a sideway sensor that can tell if there is a car or the spot is empty. Once you find the spot to park in, you simply press the park button. The side sensor tells you how far to drive forward and when to stop (something that humans have hard time gauging accurately and precisely while computers will perform this flawlessly every time). After that, the computer simply executes pre-programmed tasks like turning the wheel to one side a certain amount, then once the sensor no longer sees the car beside it, the wheel is straighten out and then turned in the opposite directions. Sensors in the back tell the computer when to stop. This is a repetitive task with very small amount of variables, something that computers excel at.



Now think how many sensors would be needed to mimic human's perception of the speed, traffic ahead, on the sides and behind. Anticipation of curves, hills and something we call situational awareness, all of which affect how the diver release the clutch and changes gears and at which point. On top of that imagine the amount of programming needed to duplicate human's response based on all these inputs. It becomes mind boggling.

I remember learning about robotics and we had Nachi and Fanuc robots to play with. Each of those arms had a computer rack the size of a dresser to run them. These robotic arms could duplicate the smooth motion of painter's arm when spraying a car. This simple motion, to us, could not be replicated for decades and it was only possible because of huge advances in computing power.

Computers, robotics and automation are really amazing fields and great strides have been made, but they don't even scratch the potential of human abilities.

 

[JetStar]

Systems that I'm familiar with do not actuate the clutch during the shift. The system just cuts ignition for a pre-determined time during the shift to unload the gears for very fast shifting.

 

[Vikas]

I am enjoying this discussion and learning a lot. I am comp sci major and my job is in computers although I only have limited exposure to robotics.

I believe I am talking about releasing the clutch portion. All other things remain the same between current automatic and computer shifting the manual transmission. I am not sure why that is being brought up repeatedly. May be my understanding of current automatic transmission is wrong but I thought the decision to select appropriate gear is done by electronics in today's automatic transmission i.e. by so called TCM after taking input from lot of sensors but mostly engine load, rpm and throttle position etc. Or is it really done by the fluid itself? I imagine that is how it must have been done fifty years ago when automatic transmission was first developed.

Today in a two pedal vehicle, the interpretation of the two pedal input drives the computer to select the appropriate gear. Clutch modulation really should not need huge computing power or any additional sensor. If you are telling me that special situational awareness is needed to release the clutch, I am calling you on it.

Algorithm (assuming throttle by wire):-
Shifting gear? Depress clutch
Only slip the clutch when absolutely necessary
Rev match (pretty easy given throttle is now controlled by the computer)
don't dump the clutch
if the engine is close to stall speed and is already in the lowest gear, if there is momentum put it in neutral; if going up the hill and speed is below engine stall speed, use pulse width modulation to control the clutch etc etc

I mean even the above algorithm would have computer driving as good as a new driver. Version 7 of this algorithm would have computer doing heel and toe downshifting

Let us discuss the clutch aspect and not the rest of the transmission as that is already being done today.