Summer of Code 2008: Motion Synthesis based character animation
Student: Du Min (westine)
Mentor: Paul Cheyrou-lagrÃ¨ze (tuan kuranes)
Status: In development
Motion DataBase The BVH Conversion Release of CMU's Motion Capture Database
This work is aim to add more features into OGREâ€™s animation subsystem.
To my knowledge, OGRE now support basic animation like skeleton and mesh vertex animation, three type of blending methods.
Fortunately, Tuan Kuranes has been done a lot of interesting and wonderful work on it already, which most attracts me is ODE add-on. Using this, I plan to add dynamic force on character as another environmental control way, which will then be treated together with other input stuff such as interactive user input, AI, target, obstacles.
For example, synthesizing believable soldiers would involve representing the stylistic walking and running motions while generating realistic dynamic responses to attacks from explosion and short, which will result in reaction and responsive actions. There is no single motion synthesis technique that can create these wonderfully realistic characters, so multiple techniques mixture is needed. Motion graphs have been extensively researched and applied in this domain ever since 2002 ago. Pros of motion graph is its best effort of preserving original motion as much as possible, which meanwhile causes its cons with larger diameter of graph size and complexity of the graph structure, which in turn makes the connectivity and search algorithm more or less intractable.
On the other hand, physical simulation involving inverse kinematic and dynamic driven character animation accurately expresses the physical interactions between a character and its environment, although the artifacts are usually observed.
The same as other problems in academic field, there is no key to every door. In practice, we can just one right for its corresponding door. This work is a hierarchical and composite animation platform. Its base is motion graph, which benefit a lot from recent motion graph extension such as parametric motion graph and interpolated motion graph. Transitions are the most annoying issue in motion synthesis whether which technique is used. Traditional motion graph takes great effort to keep transitions as smooth and natural as possible, and this is often done both with preprocessing and post-processing step besides transition blending itself.
In order to alleviate the labor consumed for work such cutting out motion clips and detecting gait cycleâ€™s foot plants, I adopt prior knowledge
learned from motion data to assist guide motion generation.
April 1 ~ April 21
Make sure which are most needed for OGRE with help from mentors and all the OGRE community.
April 21 ~ May late
Determine which style of my work will be added to OGRE, code,add-on,plug-in,samples,editor. Anyway, projects of code should be built.
This is also the period which I get familiar with the develop habit of OGRE, since recent I begin to really like to be a long time member of OGRE.
Stage 1 --- May 26 ~ June 15 [week 1ã€2ã€3](Completed):
Overview: Convert bvh format into Ogre Mesh and Skeleton, single animation can be played.
week 1 [Done] Bvh file read, data conversion to Ogre Skeleton.
week 2 [Done] Using bvh created animation track to drive new character mesh.
week 3 [Done] Convert a list of bvhs in batch
This stage cost more time than estimated due to additional skeleton and mesh binging work.
Stage 2 --- June 16 ~ June 29 [week 4ã€5]ï¼ˆCompleted)ï¼š
Overview : Porting motion graph into OGRE.
week 4 Create Motion graph Class in Ogre as a demo sample.
week 5 Combine motion graph with OgreAnimation, node and edge mapping to animationstate and animation track.
Stage 3 --- June 30 ~ July 14 [week 6ã€7]:
Overview: Build motion graph and Randomly walking in Motion graph by script
week 6 Create and connect nodes, slice and merge edges by script.
week 7 Use basic graph search algorithm to go through.
MileStone 1 --- July 1 ~ July 14:
In this checkpoint, Ogre has its skeleton character animated along a motion graph, but the motion graph is small and simple, and transition
are not smooth.
This motion graph is built from a script, the script is manually edited without gui.
All animation resources are packaged in advance not grouped as the script is interpreted.
Mid-term evaluations, modifying, fix bugs, improving.
Stage 4 --- July 14 ~ July 27 [week 8ã€9]:
Overview: Motion graph build work by script
week 8 A simple script editor gui, user can create finite state machine and assign animation to state.
week 9 Automatically build motion graph from Animation Resources following script.
Stage 5 --- July 28 ~ August 10 [week 10ã€11]:
Overview:Fade in Fade out smooth transition between animations.
week 10 Align a character's face in the same direction when two animation track are connected.
week 11 Blending the overlapped slices. For example, make running to walking transition more smoother.
MileStone 2 --- August 11 ~ September 1 [week 12ã€13ã€14]:
In this checkpoint, user can create a character animated based on a motion graph, this motion graph is built from a script.
The script is editor in a gui, user is able to assign bvh format animation to a state in this script editor.
Stage 6 --- August 11 ~ September 1 [week 12ã€13ã€14]:
Reinforce everything as possible as I can to give a satisfactory work
Import Motion Capture Data
Stage 1 (Done):
week 1 - 2
Something to decide:
Because motion capture data only has skeleton and animation, but no mesh or other material is bound within, and so far as I know, all 3D content authoring tools that has OgreExporters need mesh to work.
I propose two alternative ways to import mocap:
- As Sinbad has suggested, Automatic Rigging (http://www.mit.edu/~ibaran/autorig/) is needed if motion capture data is directly imported into Ogre without other 3D tools and their associated OgreExporter.
This is very helpful, take these words for examples "make neither head nor tail of blender" said Sinbad, "a complete novice with 3Ds Max" said another Ogre guy.
- As you can see in the above image, Collada is in insight. We can follow the Collada schema, and motion capture data may become another common 3D content material, and with OgreCollada, we can use these content in Ogre as other resource.
But for this summer code, it is too far beyond, and as Kex( the mentor of OgreCollada) said to me "I don't know if you can use this. If you change that part of the OGRE API we should merge this after GSoC, IMHO. You students should concentrate on your main work."
So this should be considered later.
After all, I must implement motion graph in Ogre as my major work, which require motion data to be imported into Ogre as quickly as possible. I would like to use an existing mesh in Ogre such as jaiqua.mesh in the Oger skeleton animation demo, all the mesh information include triangle faces, bone hierarchy , bone weight assignments will be retained, only animation in .skeleton is replaced. This is a simple way of automatic rigging.