Motion Synthesis on the GPU

Recently motion graphs were developed in order to automatically generate long sequences of smoothly changing
motions of a captured motion data. To create motion graphs, motion capture data has to be segmented at temporal
positions that allow for synthesizing motions with smooth transitions. Usually such synthesizing of motions is carried
out on a CPU and the resulting poses are transmitted per animation frame to the Graphics Processing Unit (GPU).
Since the operations involved to synthesize motions can be executed in parallel the GPU would be a more
appropriate place to compute new motions. In this project methods are developed that allow us to animate virtual
characters directly on the GPU by synthesizing motion segments without having to transfer pose data per frame
from the CPU to the GPU. Using such a GPU approach for motion synthesis leaves more processing power for higher
level artificial intelligence parts for the CPU.