Latent motion spaces for full-body motion editing

• Published in: The Visual computer 2013-03, Vol.29 (3), p.171-188
• Main Authors: Carvalho, Schubert R., Boulic, Ronan, Vidal, Creto A., Thalmann, Daniel
• Format: Article
• Language: English
• Subjects: Adaptation; Artificial Intelligence; Computer Graphics; Computer Science; Constraint modelling; Editing; Handles; Image Processing and Computer Vision; Interpolation; Inverse kinematics; Kinematics; Optimization; Original Article; Spacetime; Spline functions
• ISSN: 0178-2789; 1432-2315
• DOI: 10.1007/s00371-012-0678-z

We explore an approach to full-body motion editing with linear motion models, prioritized constraint-based optimization and latent-space interpolation. By exploiting the mathematical connections between linear motion models and prioritized inverse kinematics (PIK), we formulate and solve the motion editing problem as an optimization function whose differential structure is rich enough to efficiently optimize user-specified constraints within the latent motion space. Performing motion editing within latent motion spaces has the advantage of handling pose transitions and consequently motion flow by construction from single key-frame editing. To handle motion adjustments from multiple key-frame and trajectory constraints, we developed a latent-space interpolation technique by exploiting spline functions. Such an approach handles per-frame adjustments generating smooth animations, while avoiding the computational expense of joint space interpolations. We demonstrate the usefulness of this approach by editing and generating full-body reaching and walking jump animations in challenging environment scenarios.