The paper introduces GAMMA (Generalizable Articulation Modeling and Manipulation for Articulated Objects), a novel framework designed to model and manipulate articulated objects like cabinets and doors. These objects are common in daily life but pose significant challenges due to their diverse geometries, semantic categories, and kinetic constraints. Previous methods often focus on specific joint types or grasp poses, lacking generalizability to unseen objects. GAMMA addresses this by learning articulation modeling and grasp pose affordance from a diverse dataset of articulated objects. It employs adaptive manipulation to iteratively improve the articulation model and enhance manipulation performance. The framework is trained using the PartNet-Mobility dataset and evaluated in both simulation and real-world experiments with a Franka robot. Results show that GAMMA significantly outperforms existing methods in unseen and cross-category articulated objects, demonstrating strong generalizability and improved success rates in manipulation tasks. The contributions of GAMMA include its ability to generalize across categories, its physics-guided adaptive manipulation, and comprehensive experimental validation in both simulated and real-world environments.The paper introduces GAMMA (Generalizable Articulation Modeling and Manipulation for Articulated Objects), a novel framework designed to model and manipulate articulated objects like cabinets and doors. These objects are common in daily life but pose significant challenges due to their diverse geometries, semantic categories, and kinetic constraints. Previous methods often focus on specific joint types or grasp poses, lacking generalizability to unseen objects. GAMMA addresses this by learning articulation modeling and grasp pose affordance from a diverse dataset of articulated objects. It employs adaptive manipulation to iteratively improve the articulation model and enhance manipulation performance. The framework is trained using the PartNet-Mobility dataset and evaluated in both simulation and real-world experiments with a Franka robot. Results show that GAMMA significantly outperforms existing methods in unseen and cross-category articulated objects, demonstrating strong generalizability and improved success rates in manipulation tasks. The contributions of GAMMA include its ability to generalize across categories, its physics-guided adaptive manipulation, and comprehensive experimental validation in both simulated and real-world environments.