GaussianArt: Unified Modeling of Geometry and Motion for Articulated Objects

GaussianArt represents a breakthrough in the reconstruction of articulated objects, addressing the limitations of prior methods that often separated geometry from motion. By employing articulated 3D Gaussians, this unified approach significantly enhances robustness and scalability, allowing for the effective modeling of complex objects with up to 20 parts. The introduction of the MPArt-90 benchmark, which includes 90 articulated objects across 20 categories, provides a comprehensive framework for evaluating the performance of this method. Extensive experiments demonstrate that GaussianArt consistently outperforms previous techniques, achieving superior accuracy in part-level geometry reconstruction and motion estimation. This advancement is crucial for applications in robotic simulation and human-scene interaction modeling, as it facilitates the creation of more accurate digital twins of interactive environments.