Repurposing Hand Animation for Interactive Applications

How to read this

Category

Method: data-driven animation synthesis from an authored corpus

Contributions

• Learns separate low-dimensional embeddings of a keyframed hand-animation corpus per semantic label using a Gaussian Process Latent Variable Model mapping rig controls to a 3D latent space
• Generates novel motion by moving a simulated particle through the latent spaces, with bridges between similar poses enabling smooth transitions across semantic labels
• Demonstrates interactive character animation on the face of Toothless from How to Train Your Dragon 2 playing a game with the user

Context

Sits in the data-driven motion synthesis lineage, applying the Gaussian Process Latent Variable Model to repurpose existing keyframed rig animation for real-time interactive control.

Correctness

Validated by a single qualitative interactive demo (a production dragon face), so generality across rigs and richer interactions is suggested rather than measured, and quality depends on the coverage of the input corpus.

Clarity

Application idea is accessible on a first pass; the GPLVM latent-space formulation and bridging warrant a second pass.

How to read it

First pass for the corpus-to-latent-space framing and the bridge-transition idea; do a second pass on the GPLVM mapping and particle dynamics if you intend to reimplement or assess fidelity.