Reproduce “Action Transfer Across Character Body Plans in Latent-Action World Models: An Empirical Study of Animated Characters” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Amarnath R
The authors evaluate whether latent-action world models transfer body-appropriate motion between characters using an action-specific pixel-prediction score. They compare the correct transferred action with wrong actions from the same source. On animator-made animations, M4 scores S = 0.46 for held-out animal families and 0.20 for humans; the score measures prediction, not limb mechanics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.