This study presents a novel approach to cloth unfolding, aimed at enhancing convergence speed and efficiency in reinforcement learning applications. Initially, a self-supervised learning framework is employed to train the generation of pick-and-drag actions utilizing dual robotic arms, guided by collar features and contour analysis. The framework exhibits an 78% success rate in unfolding within three maneuvers in simulated environments. Drawing inspiration from the air-drying process of garments, a hanger is employed to exploit gravity for effective cloth flattening. Experimental validation, conducted on a dual-arm robotic system equipped with an RGB camera in real-world scenarios, validates the method’s efficacy, achieving a 95% unfolding rate post-hanging.