I'm trying to block out some time to work with ODE again and
wanted to take advantage of all the fabulous work done on XODE.

To start, I'm creating collisions between multiple car objects.
(The cars defined via XML naturally) Seems like i ought to be
able to check for collisions between the cars' container spaces,
rather than the individual geoms. (I've used the chassis in the past)

The schema doesn't look like it supports name attributes on the
space node like it does on bodies and geoms. So, I'm
confused how i'd go about getting at the NativeAddr for the spaces
so i can compare them inside iterateContacts.

The examples that show this sort of logic don't use XODE, and the
XODE examples don't use iterateContacts.

This is probably *really* easy. Anyone got some sample code
they can point me towards?

I'm a dumbass -- it says quite clearly in the docs you have to set the dContactFDir1 flag.

Ack!

contact.setFdir1 appears to have no effect?!

no matter what vector I pass, it behaves as though
Fdir1 is <1, 0, 0>

and the units for getAngularVel() ?

radians / step ?

additionally, when i try to compute the friction direction vector by hand, i'm having really weird results. Are getAngularVel and getLinearVel in body coords or world coords? Does contact.setFdir1 expect body coords or world coords?

Seems like both should be world coords, but i've been mistaken before.

:j

"body groups" would be a collection of bodies joined by joints. My guess above proved correct, i needed to move both the chassis *and* the four wheels.

I've no idea how to GetRelPointVel in javaode.

hints?

this method doesn't appear to have a high level api analog, and I'm not able to properly cast to the required types without crashing the server. Thought I'd just ask if anyone had a sample code block.

I could just use the body's rotation/linear velocity, but i thought I'd try the built-in method first.

sweet. I'll download the latest stuff this week.

I have mad respect for anybody who posts fixes at 2am on saturday night by-the-way.

the car demo basically destroys all the joints
and rebuilds the car from scratch. I see why this
is effective now. Moving the chassis and expecting
the wheels to snap into place probably creates
crazy momentums and accelerations. I wonder
if I can get away with moving the wheels but leaving
the joints alone ...

I don't supppose Ode has a notion of "body groups"
which would allow multiple arbitrary bodies to be
positioned and rotated as a group?

thanks,  :j

i changed my walls from planes to boxes, and
not only does the debug version work
(boxes are placeable, planes not so), but i
haven't crashed it since (knock on wood).

So, in my very uninformed opinion, there're problems
somewhere with planes...

:j

resetRotationAndForces mostly never works.

setting a body to a known position, orientation, etc ought to be pretty elementary operation. Why does the simulation blow up half the time? workarounds?

my car sim allows the car to flip over, so i'm implementing a reset button ...

:j

I did the same google ; )

no, no body. In fact, at the execution point at which I get the msg, the two lines of code I posted are the only references to the geom object at all.

I'm glad the debug lib gave me a message too, but it's at an execution point prior to that of my original issue, and therefore (most likely) unrelated.

The car example uses spheres instead of cylinders.
I was thinking i might make this modification -- as
much to teach myself more about ode as anything.

I can't figure out which way is "up". How are cylinders
(and cones for that matter) oriented by default?
Is this documented somewhere?

Would anyone like to post a simple example that creates
a cylinder and rotates it into a known orientation?

           GeomPlane rightwallGeom = new GeomPlane(-1f, 0f, 0f, -600f);
           space.add(rightwallGeom);


Does the fact that a GeomPlane isn't placeable have something to do with it may-haps? ; )

Not that I have any idea what to do about it...
:j

a thousand thanks!

interestingly, i have yet to crash ode with the new
DLL (EDIT: now i have!) and when I attempt to use the debug version,
i get the following error popup:

ODE INTERNAL ERROR 2
geom must be placeable (ode/src/collision_kernel.cpp:389)

This is during ODE initialization, before any clients are connected to the system, which means there are no "cars" or collisions yet -- only the 4 walls of the environment.

curious and curious-er

Source available by request.

any chance someone out there could compile me a debug DLL?

C++ isn't my bag and, being on windows, I'm having a hell of a time finding a free compiler that wants to play nice with the makefile.

TIA

I've been adapting the car example for use with unity (a multiuser server: http://www.moock.org/unity )

I get intermittent crashes in the odejava.dll

It's an EXCEPTION_ACCESS_VIOLATION *usually* in dJointAddSliderForce.

The only similar issues I can find suggest increasing the amount of memory available to the java VM, but I haven't found that to be a reliable solution.

Any thoughts?

turned out i was using an old DLL.

DOH!

thanks -- i'm moving along fairly smoothly now -- odejava is the heat.
:j

this ought to be *extremely* easy, but...

I've created a simple sphere in an environment with a little gravity.
I also have a dPlane which i use as the floor (i've also tried a large box).

No matter what settings i use (stepSize, surfaceMu, object densities, etc),
the sphere either falls through the floor immediately, or
comes to rest, stays completely still for a few seconds, and
then continues as if the floor has disappeared.

What gives? The truly baffling thing is that if I trace out the
position values every step, they're constant for many many iterations
and then all-of-a-sudden start changing again. Bizzare!

source available by request -- but this has GOT to be a common issue.