HaNDF: Object-Conditioned Geometric Neural Hand Distance Fields

This paper proposes HaNDF, an object-conditioned geometric neural hand distance field designed as a data-driven prior for modeling plausible hand-object configurations. By leveraging Riemannian geometry, the method represents hands within the zero-level set of an object-induced neural field and projects given poses onto this manifold. The authors demonstrate that HaNDF enables flexible pose refinement, reconstruction under occlusion, and physically plausible manipulation synthesis, serving as a powerful prior for embodied AI applications involving hand-object interaction.