[Opensim-users] Changed default av_capsule_tilted to false in OpenSim master

[nlin]

Hi,

2009/11/10, Justin Clark-Casey <jjustincc at googlemail.com>:
> There was a bit of discussion last week in the OpenSim dev's office hour and on
> irc that changing av_capsule_tilted = false in OpenSim.ini.example would be a
> good idea in the master branch for avatar related physics for ODE.

The av_capsule_tilted=false should generally be safe. Please let me
mention some general things to watch out for. These shouldn't be a
problem now, but were problems that I encountered and hopefully fixed:
1) Avatar falling through terrain
2) Extremely rapid changes of direction and movement could conceivably
cause the avatar capsule to become unstable.

Note that these items are refer to poor behavior in terms of the
physics engine itself (visualized with drawstuff), as opposed to ways
that poor behavior might manifest itself in the viewer (such as
bendy-knees).

Regarding 1, terrain collisions, reducing the avatar tilt seems to
increase the tendency of the avatar to fall through the terrain - this
has something to do with the ODE terrain collider, which I wasn't able
to fully debug - anyone wanting to give it a shot, let me know. As a
workaround, currently a small, experimentally-determined tilt is
maintained (even with tilt=false) to prevent falling through the
terrain. This isn't ideal, but seems to work well enough.

Regarding 2, rapid changing of avatar direction: The tilt rotates with
the avatar, snapped in 90-degree increments to one of  four directions
(NE,NW,SE,SW), so that the direction of the tilt always is opposite
the direction of movement. In other words, if your avatar is moving
forwards, the avatar is leaning slightly backwards. The reason the
tilt is rotated with the avatar is to reduce/remove
direction-dependent behavior, such as being only able to climb over
prims from a certain direction. The reason the tilt rotation is
snapped to 90-degree increments, instead of smoothly rotating with the
avatar, is to prevent extremely rapid changes in tilt orientation,
which is done by setting ODE motor stops. If you change ODE motor
stops too rapidly, the avatar capsule may "explode" and go flying
several hundred meters into the air. Conceivably, if your avatar
walked directly straddling a line facing northeast and constantly did
a zig-zag motion above and below the NE line, this could cause the
avatar orientation to change rapidly enough to cause avatar
instability. In practice I doubt that this will be an issue. (Still,
would be interesting to write a bot to try to provoke this behavior.)

More generally, however, this dynamically rotating tilt is not ideal.
I think there are 2 avatar physics issues that need further
investigation:

1. Removing the tilt completely instead of just reducing it. Currently
not workable because the terrain collider then allows the avatar
capsule to fall through sometimes. Fixing this would allow complete
consistency (direction-independence) in avatar physics behavior, close
avatar interactions like hugging, and no need for changing the motor
stops while the avatar is moving, reducing risk of explosion.

2. Allowing flexible height of steps that the avatar can step over.
Currently the bottom-most hemisphere of the avatar capsule always
drags against the terrain or any objects it collides with. This has
the limitation that the height of obstacles that can successfully be
stepped over cannot be directly tuned, as it depends indirectly on the
size of the hemisphere and the specifics of collision handling (things
like ERP, CFM, bounce). Ideally we would want to be able to directly
tune the height of "steppable" objects in OpenSim.ini, or perhaps have
this be different per avatar (dependent e.g. on the avatar height).
But it looks like we would probably have to implement a new avatar
physics mechanism to handle this flexibly: something like a floating
capsule, capsule+sphere, or using mesh collisions instead of primitive
capsules.

-nlin