Abstract
While current general-purpose 3D human models (e.g., SMPL-X) efficientlyrepresent accurate human shape and pose, they lacks the ability to physicallyinteract with the environment due to the kinematic nature. As a result,kinematic-based interaction models often suffer from issues such asinterpenetration and unrealistic object dynamics. To address this limitation,we introduce a novel approach that embeds SMPL-X into a tangible entity capableof dynamic physical interactions with its surroundings. Specifically, wepropose a "half-physics" mechanism that transforms 3D kinematic motion into aphysics simulation. Our approach maintains kinematic control over inherentSMPL-X poses while ensuring physically plausible interactions with scenes andobjects, effectively eliminating penetration and unrealistic object dynamics.Unlike reinforcement learning-based methods, which demand extensive and complextraining, our half-physics method is learning-free and generalizes to any bodyshape and motion; meanwhile, it operates in real time. Moreover, it preservesthe fidelity of the original kinematic motion while seamlessly integratingphysical interactions