[Opensim-dev] robot simulation

[dan miller]

--- Jordi <mumismo at gmail.com> wrote:

> > As far as I know, there is no simulation environment that is both
> accurate
> > in terms of physics modeling, and affordable and social at the same
> time. An
> > environment like that would be a major enabler for lots of cool stuff --
> the
> > word used these day is "transformative". 

Hear hear! This is of course more or less what is proposed in works like
Snow Crash and The Matrix (without Mr. Smith antagonizing Neo, it would have
been a cool place to hang).

I'm also thinking of
> > special-purpose, not necessarily affordable, clients for it like the
> "cave"
> > down at UCSD, a fully immersive 3D environment where you can physically
> move
> > around the objects, using special goggles -- it's such an intense
> feeling!

I know the Cave quite well -- a family friend was one of the core developers
(Tom DeFanti, big SigGraph guy).  It's an amazing thing to experience it; it
really gives you a feel for where this stuff is headed.  But long before a
cave is as common as a hot tub, you'll have 3D goggles and home user mo-cap
using webcams and bracelets.  

This sort of interface is actually quite important because the present game
paradigm of joystick/cursor control of one's avatar and menu-based
manipulation of objects is not nearly precise or immersive enough to
integrate well with a serious physics simulation.

> > Just imagine the wonders that that would do for real world systems
> > engineering! (ok, and for sexual and warlordy fantasies too, but you
> don't
> > need good physics for those :-)

Hmm... I think physics is an integral part of sexual congress, but I
digress...

> Imagine a world where entities are not controlled but programmed. In C#,
> lsl, etc. You get sensor data and you have to make your entity or entities
> win and conquer others.
> Physics and sensors information can open unimaginable new possibilities.

> The first issue I can think of is that not only animation based movement
> (not controlled by physics) but articulated based movement (controlled by
> physics) will be used. Maybe the physics based movement can be translated
> to
> animation based movement for SL compatibility.

I've said this before, but I think there are major issues with integrating a
truly accurate simulation with the animation approach of videogame design. 
There are algorithms out there that can take an animation file and transform
it into physically realistic commands to a 'robot', where robot includes
something like an articulated avatar with virtual muscles.  If you cheat and
just make the object follow the animation script, you will have non-physical
things like infinite forces and such, which can destabilize the simulation.

> 
> The second issue is that using Collada can be a good idea (physics and
> visual information in the same format), the ogrecollada project is very
> active now (openviewer will use Ogre3d, right?) and I am going to add
> collada physics support to mine soon also. I don't know how can this be
> mixed with current SL mesh format.

The SL problem is that it's prim based, and the rest of the world is
mesh-based.  We've discussed this before, and I think the consensus view is
that mesh capabilities need to exist in the viewer to do almost everything
we're talking about.  If SL doesn't want to go that way, I predict the
client will fork sooner rather than later.

Yes, but at the risk of repeating myself, the real challenge is to integrate
non-physical scripting (such as motion-captured avatar moves) with the
physics in a sane and consistent way.  This is a large part of my day
research -- trying to get a bipedal robot to do what you want it to do,
while respecting the laws of physics.  It turns out to be (almost) as
difficult to do this in a realistic simulation as it is in reality.  

What I'm getting at is, if you want your avatar to walk around in a
physically realistic simulation, you have to treat it like a robot -- you
can't just shove a .bvh file into the mix.

> Doing that work in a separated branch will not be needed if it can be
> modularized away (right thing to do, don't know how difficult it can
> become).

I believe that will prove difficult/impossible in the short term.  There are
deep issues with event management that reach their messy tentacles through
all of the code; IMSHO it will be difficult to research physics while
staying compatible with what exists today.

-dan