GaussiAnimate Proposes "Skelebones" to Rig and Animate Diverse Characters from 4D Gaussian Sequences

Researchers from Westlake University, Nanjing University, the University of Hong Kong, and Macau University of Science and Technology have introduced GaussiAnimate, a rigging and animation system accepted to SIGGRAPH Asia 2026. The system targets a long-standing tension in character animation: kinematic skeletons offer intuitive control but struggle with non-rigid surface dynamics like loose clothing and soft tissue, while free-form bones capture those deformations but lack a structured control hierarchy. GaussiAnimate bridges the gap with a "skelebones" representation that combines a Mean Curvature Skeleton with free-form bones compressed from temporally-consistent deformable Gaussians.

Animation is driven non-parametrically via a Partwise Motion Matching (PartMM) algorithm, which retrieves and blends observed deformations in response to new skeleton poses without requiring large training datasets. The authors report a 17.3% PSNR improvement over Linear Blend Skinning and a 48.4% RMSE reduction over robust LBS in motion matching, with strong generalization across clothed humans, quadrupeds, birds, and garments. Code is slated for public release for research use.