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

Christos Dimitrakakis wrote:
> the following updates are available from
> http://www.idiap.ch/~dimitrak/torcs.html
> 
> Since simuv2 is now obsolete I am updating it on CVS as well.

Well... I have set up simuv3 with your modif in order to keep
simuv2 as reference, so it's better if you update simuv3.

Bye,
Eric.
-- 
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
                        TORCS
             The Open Racing Car Simulator
         AKA The Other Release Coming Soon (Skin'r)
                  http://torcs.org
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=

Comparing the two different approaches I had for approximating the 3d
suspension:

simuv3,wheel.cpp:1.11:

	RtTrackSurfaceNormalL(&(wheel->trkPos), &normal);
	wheel->susp.x = wheel->pos.z - Zroad*normal.z;

simuv2,wheel.cpp:1.2:
	t3Dd angles;
	t3Dd normal;

	t3Dd rel_normal;
	tdble waz = wheel->steer + wheel->staticPos.az;
	RtTrackSurfaceNormalL(&(wheel->trkPos), &normal);

	angles.x = car->DynGC.pos.ax + wheel->relPos.ax;
	angles.y = car->DynGC.pos.ay;
	angles.z = car->DynGC.pos.az + waz;
	NaiveRotate (normal, angles, &rel_normal);
	wheel->susp.x = wheel->rideHeight = ( wheel->pos.z -
Zroad)*rel_normal;

The wheel->pos.z is the global model wheel position. And the Zroad is
the height of the road given that the wheel position is x,y. First of
all this is wrong because it should be (wheel->pos.z-Zroad)*normal.z;

But that still only gives us the reaction force projection to the
track normal (slightly better than normal since the reaction force can
only be perpendicular to the track surface).

However this assumes still that the car is horizontal. So in the
second version I am rotating the normal relative to the car axis. This
means that, for example, if the car is tilted at a very high angle
then the suspension calculation will only take into account the
projection of the reaction force onto the axis of the suspension
itself. (Perhaps, even though I have not added an angle for the
suspension. It could be easily added to angles though).

Now, the second model does well in limiting the push-up of the wheel
to the axis of the suspension, but what about the reaction forces that
are perpendicular to this?

I take care of that with a small hack in:

void SimWheelUpdateForce(tCar *car, int index)
{
...
	/* project the reaction force. Only wheel->forces.z is
	   actually interesting for friction. The rest just gets
	   normally at the end */
	Ft = rel_normal.x * f_z;
	Fn = rel_normal.y * f_z;
...
}

This is still not really correct, because instead of calculating the
distance between wheel->pos.z and track.Z, it would have been better
to find the point in the track surface that is closest to the wheel.
However, that is a bit complicated I think, especially for arbitrary
track surfaces. Unless of course we assume that the track is
approximately planar around the wheel. In that case, given the wheel's
position (x,y,z) and the track height Z(x,y) at position (x,y) and dZ
= z-Z(x,y) we can
calculate:
	x'=x+normal_x*dZ
	y'=y+normal_y*dZ
	z'=z+normal_z*dZ

Which is the closest point on the track surface from the wheel.
And we can also have

	dx'=normal_x*dZ
	dy'=normal_y*dZ
	dz'=normal_z*dZ

Now.. these can be rotated from the world axis to the car axis
to get dx, dy, dz. The last one will go to the suspension force
calculation. The others can be used to do an elastic collision of some
sort between the wheel and the track.

What I am doing now with the rotation - where
dz=(wheel->pos.z - Zroad)*rel_normal; - is not exactly the same as
calculating dx',dy',dz' and then rotating to get dx,dy,dz, because I
assume currently that dz' = dZ and not dz' = normal_z*dZ.

phew... That's one for the next update.

If the code works in simuv2 I'll update simuv3 later with the full
solution. However someone should patch up the source so that it works
when the car is upside down because it shoots up to infinity if that
happens.


And one more thing. Dirt-2, even when regenerated, has problems on the
first turn (segment t1), particularly at the point where t1 joints the
previous and following straights. At those points the normal is
inverted (so it points down instead of up).


-- 
Christos Dimitrakakis
IDIAP (http://www.idiap.ch/~dimitrak/main.html)

the following updates are available from
http://www.idiap.ch/~dimitrak/torcs.html

Since simuv2 is now obsolete I am updating it on CVS as well.


sounds.tar.gz:

New sounds .. 14 different engine sounds. I made a small program that
can stabilise the pitch of a given sound, so if you have any good
samples with non-steady revs, you can donate it and I'll fix it so
that it can be used in the game. (the new ferrari and lamborghini
sounds were made using this program, the source sound was a car
revving up from idle to full revs..)

simuv2.tar.gz:

aero.cpp now computes partial flow for when the airflow is cutoff
by other surfaces (only for the rear wing). Modified the flow
cut off for when you are following another car. When this happens both
your
drag and your downforce is decreased. When a car is behind you, only
the
eddies contributing to drag are reduced, so in that case I only reduce
the drag.Did the same for the wings, taking their position into
account (so the front wing will have its airflow reduced more than the
rear). Incidentally fixed a bug in the calculation (dragK=dragK*tmpas
instead of dragK=tmpas, since tmpas<1 anyway). Complaints that the
cars are difficult to drive in curves do not seem
to apply, at least not for the high-speed banked tracks now that the
full 3d
calculation is in effect.

car.cpp now calculates forces in full 3d. Note that in order to apply
accelerations we need to rotate from the car axis to the world axis,
so I apply NaiveInverseRotate() there.

wheel.cpp now calculates forces in full 3d. However, there is one
caveat
related to the suspension and reaction forces. Normally there are
reaction forces to the two axis that are perpendicular to the
suspension as well. Currently
I am approximating those in a slightly hacky way. However, tests using
an exaggerated version of dirt-2 show that the model is quite
accurate.

Other caveats: There is still a problem when the car is upside down.
Forces become unbounded and the car can shoot out to infinity.


-- 
Christos Dimitrakakis
IDIAP (http://www.idiap.ch/~dimitrak/main.html)

Hi,

The commands are:

cvs -d:pserver:ano...@cv...:/cvsroot/torcs login
cvs -z3 -d:pserver:ano...@cv...:/cvsroot/torcs co torcs

Bye,
Eric.

leolarrel wrote:
> Thank you
> 
> -----------------------------------------------------------------
> 每天都 Yahoo!奇摩
> 海的顏色、風的氣息、愛你的溫度，盡在信紙底圖
> http://tw.promo.yahoo.com/mail_premium/stationery.html
> 
> 
> -------------------------------------------------------
> This SF.net email is sponsored by: IBM Linux Tutorials.
> Become an expert in LINUX or just sharpen your skills.  Sign up for IBM's
> Free Linux Tutorials.  Learn everything from the bash shell to sys admin.
> Click now! http://ads.osdn.com/?ad_id=1278&alloc_id=3371&op=click
> _______________________________________________
> Torcs-devel mailing list
> Tor...@li...
> https://lists.sourceforge.net/lists/listinfo/torcs-devel
> 
> 


-- 
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
                       TORCS
            The Open Racing Car Simulator
        AKA The Other Release Coming Soon (Skin'r)
                 http://torcs.org
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=

Thank you

-----------------------------------------------------------------
每天都 Yahoo!奇摩
海的顏色、風的氣息、愛你的溫度，盡在信紙底圖
http://tw.promo.yahoo.com/mail_premium/stationery.html

Hi,

The network play is in low priority, so the development is going slowly.
Every developer is welcome, so if you want to contribute, the first step
for the network play is to send the "tSituation" structure over the net.
So The beginning will be structure->buffer and buffer->structure functions
allowing big endian systems to participate...
Knowing the the "tSituation" structure is complex and also has static parts
some preparatory work has to be done in the game first.
I have started a specification for the network play you can consult the
pdf version on <http://torcs.free.fr/specs/TORCSNetArch2.pdf> or if you want
to add comments the openoffice version is available here:
<http://torcs.free.fr/specs/TORCSNetArch2.sxw>

Happy new year 2004 to you all,
Eric.

Adrian Howchin wrote:
> Torcs team,
> 
> When will you introduce network play for this game? Is it in the pipes
> for the near future, or a long way off? Do you need another developer to
> put in some time here and there to help with this feature? I am
> moderately good with C/C++, plus I know about networking, sockets,
> tcp/ip etc. I work full-time, so obviously I won't be working on this
> 24/7.
> 
> Cheers,
> Adrian
> 
> 
> 

-- 
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
                        TORCS
             The Open Racing Car Simulator
         AKA The Other Release Coming Soon (Skin'r)
                  http://torcs.org
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=

Torcs team,

When will you introduce network play for this game? Is it in the pipes
for the near future, or a long way off? Do you need another developer to
put in some time here and there to help with this feature? I am
moderately good with C/C++, plus I know about networking, sockets,
tcp/ip etc. I work full-time, so obviously I won't be working on this
24/7.

Cheers,
Adrian

Waoo, great job :-) I tested and commited the patches to CVS.
very nice, thank you Bernhard.

Thanks,
Eric.

Bernhard Wymann wrote:
> Hi All
> 
> The memory leak is caused by the following code in grcarlight.cpp (so it 
> was a side-side effect of the mirror):
> 
> ssgVtxTableCarlight    *clight;
> ...
> clight = (ssgVtxTableCarlight*) 
> theCarslight[car->index].lightArray[i]->clone(SSG_CLONE_GEOMETRY);
> if (!disp) {
>     clight->setOnOff(0);
>     continue;
> }
> 
> If the (!disp) is taken the loop continues (->goes immediately back to 
> the guard) and the object on the heap gets forgotten. If the other code 
> path is taken, the object is inserted into the scene graph and has the 
> opportunity to become cleaned up later.
> 
> The suggested fix:
> 
> if (!disp) {
>     continue;
> } else {
>     clight = (ssgVtxTableCarlight*) 
> theCarslight[car->index].lightArray[i]->clone(SSG_CLONE_GEOMETRY);
> }
> 
> I'm not sure if that is really the intension, could you check that, Eric?
> 
> 
> 
> Fix for the virtual Destructor warning:
> 
> Add a virual destructor to grcam.h:
> 
> virtual ~cGrCamera() {};
> 
> According to the Stroustrup this is needed because the compiler cannot 
> determine the size of the object if you destruct the object with a 
> pointer to a superclass, e. g.
> 
> class manager : public employee {
> ...
> }
> 
> void f()
> {
>     employee* p = new manager;
>     delete p;
> }
> 
> So the the destructor must be virtual to call ~manager(). Eric, could 
> you check the patches, they apply to the current cvs. Thank you, bye
> 
> Bernhard.
> 
> 


-- 
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
                        TORCS
             The Open Racing Car Simulator
         AKA The Other Release Coming Soon (Skin'r)
                  http://torcs.org
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=

Hi Christos

> I would appreciate some comments.. the sources are available from
> http://www.idiap.ch/~dimitrak/torcs.html - there won't be any more
> updates until net year.

I will have a look at it on monday, but I'm sure it is great:-) Thank 
you, bye

Bernhard.

-- 

visit my homepage http://www.berniw.org

Hi All

The memory leak is caused by the following code in grcarlight.cpp (so it 
was a side-side effect of the mirror):

ssgVtxTableCarlight	*clight;
...
clight = (ssgVtxTableCarlight*) 
theCarslight[car->index].lightArray[i]->clone(SSG_CLONE_GEOMETRY);
if (!disp) {
	clight->setOnOff(0);
	continue;
}

If the (!disp) is taken the loop continues (->goes immediately back to 
the guard) and the object on the heap gets forgotten. If the other code 
path is taken, the object is inserted into the scene graph and has the 
opportunity to become cleaned up later.

The suggested fix:

if (!disp) {
	continue;
} else {
	clight = (ssgVtxTableCarlight*) 
theCarslight[car->index].lightArray[i]->clone(SSG_CLONE_GEOMETRY);
}

I'm not sure if that is really the intension, could you check that, Eric?



Fix for the virtual Destructor warning:

Add a virual destructor to grcam.h:

virtual ~cGrCamera() {};

According to the Stroustrup this is needed because the compiler cannot 
determine the size of the object if you destruct the object with a 
pointer to a superclass, e. g.

class manager : public employee {
...
}

void f()
{
	employee* p = new manager;
	delete p;
}

So the the destructor must be virtual to call ~manager(). Eric, could 
you check the patches, they apply to the current cvs. Thank you, bye

Bernhard.

-- 

visit my homepage http://www.berniw.org

Hi All

I have a possible fix for the memory leak and a correct fix for the
compiler warning. I will send patches and problem a description within
one hour.

bye, Bernhard.

-- 

visit my homepage http://www.berniw.org

On Fri, 19 Dec 2003, Bernhard Wymann wrote:

> Hi all
>
> >> It is also doing something with opengl, but I don't know whether these
> >> calls can lead to leaks.
>
> It's just executed once or twice, so this might be a leak, but probably
> not the big one.
>

The same thing happens with me sometimes when I change the mode from
window mode to full-screen mode. Or, sometimes when I resize the
window. Maybe it is related.

-- 
Christos Dimitrakakis
IDIAP (http://www.idiap.ch/~dimitrak/main.html)

I made some more changes to the simulation update.. with preliminary
support for full 3d. The changes are in wheel.cpp, car.cpp and
aero.cpp.

aero.cpp:

Now has lateral and bottom/top drag.

In addition to correct laminal flow using angle of attack, I also
included a factor for smoothly cuting off the flow in the rear wing
when the wind is coming from under the car.

wheel.cpp, car.cpp:

No limit on car rotation.

Forces and acceleration are computed in full 3d.
Wheel traction reaction is calculated in full 3d, with support for
cambers.

Also modified the wheel->susp.x calculation in wheel.cpp, though the
support is not complete so this feature is turned off. I also need to
add the appropriate angles for the suspension.

Caveats: All wheel forces are assumed to occur on the tread. This is
obviously incorrect in some cases. There is also no support for
collisions of other parts of the car with the track, which is
obviously very important.

I would appreciate some comments.. the sources are available from
http://www.idiap.ch/~dimitrak/torcs.html - there won't be any more
updates until net year.



-- 
Christos Dimitrakakis
IDIAP (http://www.idiap.ch/~dimitrak/main.html)

Hi all

Bernhard Wymann wrote:
> Hi all
> 
>> I think it is the mirror camera, which is the only one that allocates
>> something. Hm.. there is the fact that it does this in setViewport()
>>     if (viewCam) {
>>     delete viewCam;
>>     }
>>     viewCam = new cGrOrthoCamera (screen, x,  x + w, y, y + h);
>>
>> It is also doing something with opengl, but I don't know whether these
>> calls can lead to leaks.

It's just executed once or twice, so this might be a leak, but probably
not the big one.

bye, Bernhard.

-- 

visit my homepage http://www.berniw.org

Hi all

> I think it is the mirror camera, which is the only one that allocates
> something. Hm.. there is the fact that it does this in setViewport()
>     if (viewCam) {
> 	delete viewCam;
>     }
>     viewCam = new cGrOrthoCamera (screen, x,  x + w, y, y + h);
> 
> It is also doing something with opengl, but I don't know whether these
> calls can lead to leaks.

I don't think so (just if the OpenGL driver leaks itself).

> (Also it is the only one with a destructor which gcc
> complains cause it's not virtual, but maybe that's not related unless
> it gets destroyed/recreated at every frame)

That is also my favourite candidate:-) Perhaps the leak is caused by the 
super-class data, which becomes not destroyed. I have to look into the 
"Stroustrup".

bye, Bernhard.


-- 

visit my homepage http://www.berniw.org

On Fri, 19 Dec 2003, Bernhard Wymann wrote:

> Hi All
>
> ok, I did also a practice session with berniw 1:
> - outside camera: memory usage does not grow.
> - inside camera: memory usage grows.
>
> So it seems to be the camera.
>

I think it is the mirror camera, which is the only one that allocates
something. Hm.. there is the fact that it does this in setViewport()
    if (viewCam) {
	delete viewCam;
    }
    viewCam = new cGrOrthoCamera (screen, x,  x + w, y, y + h);

It is also doing something with opengl, but I don't know whether these
calls can lead to leaks.


(Also it is the only one with a destructor which gcc
complains cause it's not virtual, but maybe that's not related unless
it gets destroyed/recreated at every frame)


-- 
Christos Dimitrakakis
IDIAP (http://www.idiap.ch/~dimitrak/main.html)

Hi All

ok, I did also a practice session with berniw 1:
- outside camera: memory usage does not grow.
- inside camera: memory usage grows.

So it seems to be the camera.

bye, Bernhard.

-- 

visit my homepage http://www.berniw.org

Hi All

> I like to play  a lot of lap . 10 ~20 lap for wheel1,
> CG 3 , alpine 1, and more ..
> some time screen become slow or jump or X crach after
> I play some lap .

> Some day after I  detect my memory have been eat 85%
> by torcs (look : ps avx) , and if I play long race
> road , occer fast , if I play shout race road , occer
> slow.
>
> Is it bug ?

ok, I'm investigating the problem at the moment. You can reproduce it
this way:
- Choose a practice session and seclet the "Player" and alpine-1.
- Run the session.
- You don't need to drive, just let the car stay with inside camera and
mirror on, accelerate time to 4x.
- Now have a coffee and when you come back the memory usage has grown.

I can not reproduce it the following way:
- Endurance Race with 19 Robots on alpine-1.
- Outside camera.

From the above investigations I guess the fault must be in the human
driver or in the inside camera view. I suggest to have a look into the
human driver first.

bye, Bernhard.

-- 

visit my homepage http://www.berniw.org

Hi Leolarrel

> I like to play  a lot of lap . 10 ~20 lap for wheel1,
> CG 3 , alpine 1, and more ..
> some time screen become slow or jump or X crach after
> I play some lap .
> 
> Beginning,I think that is a small problem . so restart
> X , then rerun torcs  to play.
> 
> but problem alway occer.
> 
> Some day after I  detect my memory have been eat 85%
> by torcs (look : ps avx) , and if I play long race
> road , occer fast , if I play shout race road , occer
> slow.
> 
> Is it bug ?

I think so, I experienced the same problem yesterday with the current
CVS on alpine-1, endurance race with 20 cars (Linux (SuSE 8.2), GeForce
(driver 4496), 1GB RAM/2GB Swap). My system did not crash because I have
enough memory.

bye, Bernhard.

-- 

visit my homepage http://www.berniw.org

I am testing torcs-1.2.2-pre4

My system is Mandrake 9.1 , AMD Dulon 700M , 384M
memory + 64M swap.

I like to play  a lot of lap . 10 ~20 lap for wheel1,
CG 3 , alpine 1, and more ..
some time screen become slow or jump or X crach after
I play some lap .

Beginning,I think that is a small problem . so restart
X , then rerun torcs  to play.

but problem alway occer.

Some day after I  detect my memory have been eat 85%
by torcs (look : ps avx) , and if I play long race
road , occer fast , if I play shout race road , occer
slow.

Is it bug ?

-----------------------------------------------------------------
每天都 Yahoo!奇摩
海的顏色、風的氣息、愛你的溫度，盡在信紙底圖
http://tw.promo.yahoo.com/mail_premium/stationery.html

Hi,

In fact the PLIB author allows only one target
directory for the library to avoid this kind of problem.
but you can set CXXFLAGS and the configure script
will use it.

Bye,
Eric.

leolarrel wrote:
> Hi.
> 
> I play torcs have been 6 month.I like race car game
> very very very very mush .(beacuse I no money to buy
> race car).torcs 1.2.2 great then torcs 1.2.1.
> Yesterday I download torcs 1.2.2-pre4,It is cool! It
> have been like some PS2 or XBox race game.Thank you,
> torcs team.
> 
> About configure script of compile procsss,I have a
> problem.Torcs need plib. I dowload plib,compile and
> install. I prefer install tgz package in my package
> directory more than configure default directory.I
> install plib in /usr/game/xxx ,next I compile torcs ,
> configure script of torcs tall me it can not found
> plib,beacuse of configure only serch /usr/include and
> /lib.So I could only link /usr/game/xxx to
> /usr/include . I think our should can add a
> -with-plib= param in configure script. do this , gcc
> will add -I(plib include) -L(plib library).Another our
> should can add check env args CFLAS and CXXFLAGS
> function in configure.
> 
> I use MesaGLUT ,so I have not get same problem about
> fglut. But I think that is a some status.
> 
> Thank you, torcs team.
> 
> 
> -----------------------------------------------------------------
> 每天都 Yahoo!奇摩
> 海的顏色、風的氣息、愛你的溫度，盡在信紙底圖
> http://tw.promo.yahoo.com/mail_premium/stationery.html
> 
> 
> -------------------------------------------------------
> This SF.net email is sponsored by: SF.net Giveback Program.
> Does SourceForge.net help you be more productive?  Does it
> help you create better code?  SHARE THE LOVE, and help us help
> YOU!  Click Here: http://sourceforge.net/donate/
> _______________________________________________
> Torcs-devel mailing list
> Tor...@li...
> https://lists.sourceforge.net/lists/listinfo/torcs-devel
> 
> 


-- 
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
                       TORCS
            The Open Racing Car Simulator
        AKA The Other Release Coming Soon (Skin'r)
                 http://torcs.org
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=

Hi.

I play torcs have been 6 month.I like race car game
very very very very mush .(beacuse I no money to buy
race car).torcs 1.2.2 great then torcs 1.2.1.
Yesterday I download torcs 1.2.2-pre4,It is cool! It
have been like some PS2 or XBox race game.Thank you,
torcs team.

About configure script of compile procsss,I have a
problem.Torcs need plib. I dowload plib,compile and
install. I prefer install tgz package in my package
directory more than configure default directory.I
install plib in /usr/game/xxx ,next I compile torcs ,
configure script of torcs tall me it can not found
plib,beacuse of configure only serch /usr/include and
/lib.So I could only link /usr/game/xxx to
/usr/include . I think our should can add a
-with-plib= param in configure script. do this , gcc
will add -I(plib include) -L(plib library).Another our
should can add check env args CFLAS and CXXFLAGS
function in configure.

I use MesaGLUT ,so I have not get same problem about
fglut. But I think that is a some status.

Thank you, torcs team.


-----------------------------------------------------------------
每天都 Yahoo!奇摩
海的顏色、風的氣息、愛你的溫度，盡在信紙底圖
http://tw.promo.yahoo.com/mail_premium/stationery.html