Re: [Torcs-users] (Patch) Automatic gear shift

[Bernhard W. <be...@bl...>]

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
>