 |
|
How its done
| My most basic goal with this project is to
add looking frames so player models don't need to
tilt, therefore its the first thing I've
animated. I've also come to the conclusion that
short of true skeletal animation the id-md3 model
scheme of splitting players into torso and lower
body is best. It allows looking whilst running,
standing, crouching, and swimming and gives the
most animation freedom (of any split model
solution w/o in-game boning). And while we're at
it I'll divide the head as well to maximize skin
coverage and allow skin artists to crank that
facial detail, and swap heads. Back to looking, |
|
Here we have my new test model aiming the Q2 blaster up to about 80 degrees give or take, the incriments are ten degrees between each frame. Hence, one should never be more than 5-6 degrees away from the player's "true" viewing angle (looking strait up or strait down are exceptions). Looking down is the same and plus eight more frames that makes seventeen aiming frames. I should mention that we'll only be representing degrees of a player's view pitch and not angle because the angle can be handled in-game. |
| Okay, so we can get the dude looking up and
down, why bother? First and foremost it allows
animation mixing between the upper and lower
bodies, so now when a player's lower body is
running, jumping, swimming, or crouching their
upper body can be looking left or right and
aiming up or down. This also reduces the amount
of data per model because the lower body doesn't
need 17 frames to look around, only a simple
standing (aggressive) frame. The lower body
doesn't have to store swimming, jumping, or
crouching because its pose remains unchanged.
Once the critical issue of breaking the animation
sync between modelindexes (aka tris.md2 the
player and his attachments like weapons or flags)
is fixed, a larger issue is how to align these
separate models during some radically different
poses. Quake 3 Arena, the boxed version, uses special triangles in a model known as tags to align the torso, head and legs (in the test they all shared the same model origin). But since I'd like to avoid messy formatting issues I'll be using a solution proposed by Fugitive on the RUST forums. His idea is to store offset points in a txt file and keep the file in the baseq2\player folder with these enhanced models. Then as the models are loaded, the txt file would be read and these offset values used.
For the alpha release of this test model I'm using offset "states" which get stored once and apply to one or more animations. There is only one offset at the moment, its unlikely there will be more, and its used to position the upper body (and its derivative models like guns, flags, the head, etc.) whilst the lower body uses the traditional Q2 player model origin; the one that sits 24 units off the ground. Cerain animations, such as death, involve the whole body and as such a null offset is used and all parts of the model must be positioned at the traditional origin and appear in frame sync. For the alpha release there are only 3 states (aside from null), one for standing, one for the animation called runTestB and a final state for the crouched pose. Null offset animations include runTestC, crawlTest, and all but the first frame of jumpTest. Redundant frames between the models have been eliminated so the upper body excludes the 1st frame of jumpTest, all of runTestA, and runTestB. The lower body doesn't include any aiming frames, holstering, switching, or throwing frames. If you haven't already then check out the text file included with the alpha release (521K). QDevels thread dicussing implimentation Questions, comments, insults,
all welcome and appreciated! |
&
Microsoft's Front Page Express