[ODE] movement force conversion

Piotr Obrzut piotr_obrzut at o2.pl
Wed Mar 16 17:18:37 MST 2005


Hi,

15 marca 2005 (23:15:40) Jon wrote:



JW> The physics system will ensure that the feet remain planted, because
JW> it will create contact joints between floor and foot, and you give it
JW> some amount of friction. If you have a sloppy animation, it will 
JW> exceed foot friction, and it will slip. If you make friction low, it
JW> will "walk" as if on ice, and quite possibly fall over. The draw-backs
JW> of a physically modeled system!

JW> You're right in that the entire leg adds inertia that makes the hip
JW> bending harder than it "should" be. Similarly, when bending the hip
JW> bone, you actually also counter-bend the pelvis it's attached to. 
JW> However, if you accumulate error, and increase the torque applied for
JW> errors that remain after a few steps (a la a typical integrating 
JW> controller system), it will all average out. To make the character not
JW> fall over, either be really good with your animation, or add a magic
JW> torque to the root bone (typically, your pelvis) that attempts to 
JW> right it up (make it match the animation) every frame; again with an
JW> error integration to yield extra force as necessary.

JW> I know it sounds somewhat crazy, but it really does work, assuming
JW> you have a reasonable physical skeleton and reasonable walk cycle 
JW> animation.

Did you do some tests on this?

I wonder if instead of some error reducing sys. you could use this
ugly cheat:

{
at first time step add apropriate forces

check whether sth external acts on your model //sth external means sth
                                              //more than ground friction

if not: ignore dynamic output position and set them to apropriate
        kinematic animation pos

else : let the animation be driven by precomputed forces and start
       real dynamic simulation

}

why it could be better:
You don't have to care about error corection and it could still work
nice b/c the only reason to use this forces driven model is how your
model acts on the outside world not for handling your model. Also you
can precompute this forces offline and connect them to each anim key frame.

disadvantages:
model output forces may not be as good as with Jon Watte method.

Sorry If this is stupid but I don't have time for better reserch but
I'm very interested in that topic and I will try both methods (unless
somebody will convince me if one of them is unworthy) and then
I will report back with some examples.


-- 
greetings,
 Piotr Obrzut



More information about the ODE mailing list