ScrewNet: Category-Independent Articulation Model Estimation From Depth Images Using Screw Theory

Robots in human environments will need to interact with a wide variety of\narticulated objects such as cabinets, drawers, and dishwashers while assisting\nhumans in performing day-to-day tasks. Existing methods either require objects\nto be textured or need to know the articulation model category a priori for\nestimating the model parameters for an articulated object. We propose ScrewNet,\na novel approach that estimates an object's articulation model directly from\ndepth images without requiring a priori knowledge of the articulation model\ncategory. ScrewNet uses screw theory to unify the representation of different\narticulation types and perform category-independent articulation model\nestimation. We evaluate our approach on two benchmarking datasets and compare\nits performance with a current state-of-the-art method. Results demonstrate\nthat ScrewNet can successfully estimate the articulation models and their\nparameters for novel objects across articulation model categories with better\non average accuracy than the prior state-of-the-art method. Project webpage:\nhttps://pearl-utexas.github.io/ScrewNet/\n