<ul><li>A new approach for cross-modal biomechanical motion generation is proposed in this arXiv paper.</li><li>The method aligns latent representations of observed joint angles and ground reaction forces to denoise and disambiguate each modality.</li><li>The approach leverages the fact that local time windows of joint angles and ground reaction forces represent the same phase of the underlying dynamical system.</li><li>Experimental results demonstrate that aligning local latent dynamics across modalities improves generation fidelity and representations.</li></ul>

Cross-Modal Diffusion for Biomechanical Dynamical Systems Through Local Manifold Alignment

Discover more