torcs-users Mailing List for TORCS - The Open Racing Car Simulator

Hi David
> I will check out the r1-3-1 branch (testing will take several days).
> Also I'll look into the quaternion for car body rotation that you
> mentioned, because it looks interesting.

In simuv3 it is done already, including taking some calculations from 
"2.5D" to 3D, but it requires careful review and integration back into 
simuv2. Beware, I still want to have the DOF (degree of freedom) 
limited, because flipping over cars is no use case in a racing simulation.

> The patches show my attempts to make TORCS better; the good patches
> get used, the others are for conversation :)

Avoid thinking in good and bad, take it more as different, but valid 
point of views;-)

> The following are my qualitative observations of random TORCS stuff:
>
> I was looking into the moment of inertia, still am, and here is my
> latest comment: "max steer speed" feels like it augments the moment
> of inertia. My favorite value is 12 deg/s; this does not make the
> steering slow, it seems to filter out "twitchy" steering. Press 4
> on the keyboard while driving and look at the steering forces vector.
> car4-trb1
> <section name="Steer">
> <attnum name="max steer speed" unit="deg/s" min="1" max="360" val="360"/>

If you use a steering wheel this should not matter, because you cannot 
move it as fast as a gamepad stick, otherwise you can adjust it to your 
needs...

> In torcs/src/drivers/human, after some calculations:
> car->_steerCmd = leftSteer + rightSteer;
> Is it easy to make a purely linear steering formula that puts
> linear mouse movements into linear steering (rack and pinion)?
> I need some hints to get me on the right track.

This was confusing for me at the beginning too, but it is caused by the 
fact, that you can assign a different control for left and right steer.

> I read somewhere that TORCS determines the engine speed from the wheel
> speed from the car speed. I've noticed that the engine loses rpm while

It is not that simple, the whole thing is a connected system, I think it 
is modeled well. The only thing not covered from my point of view in the 
simuv2 drivetrain is the energy conservation when the clutch closes. 
There is an attempt for that in simuv3 too (needs as well careful review 
before integration into simuv2).

For drifting you might adjust mostly the tire properties and increase 
the ride height, e.g. read this post here: 
http://sourceforge.net/p/torcs/mailman/message/30024487/

Kind regards

Bernhard

Hi Bernhard,

I will check out the r1-3-1 branch (testing will take several days).
Also I'll look into the quaternion for car body rotation that you
mentioned, because it looks interesting.

The patches show my attempts to make TORCS better; the good patches
get used, the others are for conversation :)

The following are my qualitative observations of random TORCS stuff:

I was looking into the moment of inertia, still am, and here is my
latest comment: "max steer speed" feels like it augments the moment
of inertia. My favorite value is 12 deg/s; this does not make the
steering slow, it seems to filter out "twitchy" steering. Press 4
on the keyboard while driving and look at the steering forces vector.
car4-trb1
<section name="Steer">
<attnum name="max steer speed" unit="deg/s" min="1" max="360" val="360"/>

In torcs/src/drivers/human, after some calculations:
car->_steerCmd = leftSteer + rightSteer;
Is it easy to make a purely linear steering formula that puts
linear mouse movements into linear steering (rack and pinion)?
I need some hints to get me on the right track.

I read somewhere that TORCS determines the engine speed from the wheel
speed from the car speed. I've noticed that the engine loses rpm while
sliding sideways as a function of car angle and speed. Maybe the car
would drift better without the function of car angle affecting RPM.

Sincerely,
David Savinkoff

Hi David

Partly applied in CVS. Skipped:
- Changes in raceengine (not useful)
- Changes regarding auto reverse (would not work this way)

The other parts are in the CVS with some modifications/fixes. I cannot 
feel a difference when driving, but I do usually not drift.

You can check out the r1-3-1 branch and retest.

Kind regards

Bernhard

On 10/26/2013 05:27 AM, David Savinkoff wrote:
> Bernhard Wymann wrote:
>> Hi David
>>
>> Thank you for the patch, it looks reasonable and I will take a note to
>> put it into 1.3.6.
>>
>> Best regards
>>
>> Bernhard
>>
>> On 10/09/2013 08:30 AM, David Savinkoff wrote:
>>> Hi Bernhard,
>>>
>>> This patch improves human automatic gear shifting, making controlled
>>> drifting in TORCS much better. TORCS currently shifts gears by using
>>> car->_speed_x (a single component of a 3D vector). This is correct
>>> only when the car is traveling in the direction that is pointed,
>>> not when drifting or flying off-angle. Loss of control can no longer
>>> be attributed to improper automatic down-shifting while drifting.
>>>
>>> BTW using x, y and z vector components work better than just x and y
>>> (I drove various tracks for the past day to test TORCS).
>>>
>>> Note that this patch is more-or-less a hack-ish proof of concept.
>>>
>>> TORCS has been my high velocity entertainment for four years.
>>>
>>> Thanks,
>>> David
>>>
>
> Hi Bernhard,
>
> Here is a new patch that incorporates my previous gear-shifting
> patch in a more generalized way, allowing for the speed calculation
> to be used throughout torcs (not only for gear shifting).
>
> I applied the speed calculation to steering for an improvement too.
> (I got a hint from speed-dreams for this one)
>
> Note the gear shifting improvement and steering improvement in:
> torcs-1.3.5-test1/src/drivers/human/human.cpp
>
> The speed calculation is also applied for improvements in:
> torcs-1.3.5-test1/src/libs/raceengineclient/raceengine.cpp
> torcs-1.3.5-test1/src/modules/graphic/ssggraph/CarSoundData.cpp
> torcs-1.3.5-test1/src/modules/graphic/ssggraph/grskidmarks.cpp
> torcs-1.3.5-test1/src/modules/graphic/ssggraph/grsmoke.cpp
>
> The speed calculation could also be applied for improvements in:
> torcs-1.3.5-test1/src/drivers
>
> The speed calculation is made in:
> torcs-1.3.5-test1/src/interfaces/car.h
> torcs-1.3.5-test1/src/modules/simu/simuv2/car.cpp
>
> A constant for the yaw, pitch and roll inertial moments were
> experimented with, and changed for an improvement in:
> torcs-1.3.5-test1/src/modules/simu/simuv2/car.cpp
> Maybe these values could be made user configurable.
>
> All of the improvements make for a better driving experience.
>
> Please try this patch. I developed these patches on a modified
> car4-trb1, and tested them on car5-trb1 afterwards.
>
> car5-trb1 works excellently, and I recommend you do some
> drift-passing on corners around the robots. Now you can
> realistically scramble to save your bacon after pushing it
> too far.
>
> Sincerely,
> David Savinkoff
>
> ps.  :)
>
> If the 12.0 constant used in the Inverse Moment of Inertia is
> mathematically derived and not negotiable, here is my argument:
>
> I tried 6.0 and found that the moment of inertia was too high,
> and 18.0 found it too low (subjectively tested by driving).
>
> Thus, 12.0 must be the best value since it is half-way-between.
>
> However, half-way-between seems to have the worst of too-high
> and too-low. It is perfectly-bad.
>
> So, maybe halfway between perfectly-bad and too-something is
> the answer...
>
> 15.0 is halfway between perfectly-bad (12.0) and too-low (18.0),
> and it seems to be approximately the best value.
>
>
> The Derivation of 15.28  :)
>
> Given that 12.0 is the mathematically correct constant
> for this inverse moment of inertia formula:
>
> Constant for Moment of Inertia = 1/12
>
> 1/12 * pi/4 = 0.06545 ; pi/4 [rads] works good and looks good too!
>
> 1/.06545 = 15.28
>

Hi David

I am currently looking into it (finally). I will probably apply the gear 
shifting part and the "speed" value (because it is used in  several 
places, so it probably makes sense).

I will not change the "I" though (12.0). You can adjust the inertia 
around the most important axis with the mass repartition coefficient 
from the xml. At some point it could make sense to make this 
configurable around all axis.

But first another improvement has to be done:
- Backport from simuv3/review/implement the use of a quaternion for car 
body rotations (will help at high bankings/steep slopes etc.), google 
for things like "quaternion rotation matrix gimbal lock".

There is another thing which is not perfectly handled, see 
SimCarCollideZ, but this is sort of "no use case", namely when the car 
body hits the floor very hard (no use case because in reality this just 
trashes the car).

Kind regards

Bernhard

On 10/26/2013 05:27 AM, David Savinkoff wrote:
> Bernhard Wymann wrote:
>> Hi David
>>
>> Thank you for the patch, it looks reasonable and I will take a note to
>> put it into 1.3.6.
>>
>> Best regards
>>
>> Bernhard
>>
>> On 10/09/2013 08:30 AM, David Savinkoff wrote:
>>> Hi Bernhard,
>>>
>>> This patch improves human automatic gear shifting, making controlled
>>> drifting in TORCS much better. TORCS currently shifts gears by using
>>> car->_speed_x (a single component of a 3D vector). This is correct
>>> only when the car is traveling in the direction that is pointed,
>>> not when drifting or flying off-angle. Loss of control can no longer
>>> be attributed to improper automatic down-shifting while drifting.
>>>
>>> BTW using x, y and z vector components work better than just x and y
>>> (I drove various tracks for the past day to test TORCS).
>>>
>>> Note that this patch is more-or-less a hack-ish proof of concept.
>>>
>>> TORCS has been my high velocity entertainment for four years.
>>>
>>> Thanks,
>>> David
>>>
>
> Hi Bernhard,
>
> Here is a new patch that incorporates my previous gear-shifting
> patch in a more generalized way, allowing for the speed calculation
> to be used throughout torcs (not only for gear shifting).
>
> I applied the speed calculation to steering for an improvement too.
> (I got a hint from speed-dreams for this one)
>
> Note the gear shifting improvement and steering improvement in:
> torcs-1.3.5-test1/src/drivers/human/human.cpp
>
> The speed calculation is also applied for improvements in:
> torcs-1.3.5-test1/src/libs/raceengineclient/raceengine.cpp
> torcs-1.3.5-test1/src/modules/graphic/ssggraph/CarSoundData.cpp
> torcs-1.3.5-test1/src/modules/graphic/ssggraph/grskidmarks.cpp
> torcs-1.3.5-test1/src/modules/graphic/ssggraph/grsmoke.cpp
>
> The speed calculation could also be applied for improvements in:
> torcs-1.3.5-test1/src/drivers
>
> The speed calculation is made in:
> torcs-1.3.5-test1/src/interfaces/car.h
> torcs-1.3.5-test1/src/modules/simu/simuv2/car.cpp
>
> A constant for the yaw, pitch and roll inertial moments were
> experimented with, and changed for an improvement in:
> torcs-1.3.5-test1/src/modules/simu/simuv2/car.cpp
> Maybe these values could be made user configurable.
>
> All of the improvements make for a better driving experience.
>
> Please try this patch. I developed these patches on a modified
> car4-trb1, and tested them on car5-trb1 afterwards.
>
> car5-trb1 works excellently, and I recommend you do some
> drift-passing on corners around the robots. Now you can
> realistically scramble to save your bacon after pushing it
> too far.
>
> Sincerely,
> David Savinkoff
>
> ps.  :)
>
> If the 12.0 constant used in the Inverse Moment of Inertia is
> mathematically derived and not negotiable, here is my argument:
>
> I tried 6.0 and found that the moment of inertia was too high,
> and 18.0 found it too low (subjectively tested by driving).
>
> Thus, 12.0 must be the best value since it is half-way-between.
>
> However, half-way-between seems to have the worst of too-high
> and too-low. It is perfectly-bad.
>
> So, maybe halfway between perfectly-bad and too-something is
> the answer...
>
> 15.0 is halfway between perfectly-bad (12.0) and too-low (18.0),
> and it seems to be approximately the best value.
>
>
> The Derivation of 15.28  :)
>
> Given that 12.0 is the mathematically correct constant
> for this inverse moment of inertia formula:
>
> Constant for Moment of Inertia = 1/12
>
> 1/12 * pi/4 = 0.06545 ; pi/4 [rads] works good and looks good too!
>
> 1/.06545 = 15.28
>
>
>
> ------------------------------------------------------------------------------
> October Webinars: Code for Performance
> Free Intel webinars can help you accelerate application performance.
> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
> the latest Intel processors and coprocessors. See abstracts and register >
> http://pubads.g.doubleclick.net/gampad/clk?id=60135991&iu=/4140/ostg.clktrk
>
>
>
> _______________________________________________
> Torcs-users mailing list
> Tor...@li...
> https://lists.sourceforge.net/lists/listinfo/torcs-users
>