[ODE] IK character place feet on the ground/dynamic objects
Martijn Buijs
buijs512 at planet.nl
Thu Nov 2 06:34:40 MST 2006
Jon Watte (ODE) wrote:
>
> kxxl at poczta.onet.pl wrote:
>
>> but I don't know how they done this with genetic algoritm - isn't it too slow for real-time computation?
>
> I believe the algorithm parameters are derived genetically, but the
> actual running just uses the derived parameters. I believe they showed
> this at some GDC, perhaps five years back. They had a set of legs that
> started to learn how to walk, and they measured energy usage, and did
> genetic selection until they had a very efficient walk cycle -- only to
> realize that artists make walk cycles already, so they should apply
> their technology to something more interesting :-)
>
> Cheers,
>
> / h+
Back in 1997 the PC game Trespasser did have something similar. It worked relatively well on bipedal
dinosaurs in the game. Creatures would occasionally fall, and trip over other objects (including
their own limbs), but they reacted convincing to variable terrain and other (even moving) ground
surfaces.
When walking down-hill the walk cycle of each leg would 'pause' in fully forward-stretched pose
until it hit a surface (this also worked for jumping/falling/rolling). When walking up-hill the feet
would hit the ground 'early' in the cycle but the muscle forces pushed the rest of the body up.
They didn't seem to do any planning/ray testing where to put the feet, which wasn't a large issue as
the creatures did have an unusually good object avoidance mechanism.
The also seemed to be a stabilizing/damping force on the torso, presumably to keep it upright and
thereby avoid various unwanted behavior, only allowed the creatures to roll/trip in situations where
desirable (dead, tranquilized or injured mode).
It looked quite nice when the creatures where interacting with non static objects, or walking around
obstacles, stepping on/over objects. Straight walk cycles on flat terrain did look artificial and
'forced' (too well paced). The torso seemed to hover and the legs sort of following (neck/head and
tail movement looked more convincing). The feet where lifted unrealistically high in the walk cycle
to avoid tripping.
Most of the issues where due to lack of fine tuning (the game wasn't really finished upon release).
Perhaps random forces to slightly unbalance the creatures could have reduced the synthetic feel. The
stiffness/simplicity of the ragdoll/IK rig and the low detail of the visible geometry probably also
contributed to these problems.
If interactive character dynamics (or highly interactive player-world physics) interests you, give
the demo a try. Trespasser's (physics-)engine did some really really clever things and was well
ahead of it's time. It did some things I have yet to see in modern games. :)
Martijn
More information about the ODE
mailing list