The paper introduces ADMap, an Anti-Disturbance Map construction framework designed to improve the quality of vectorized HD maps in autonomous driving. The framework addresses the issue of point sequence jitter and jaggedness in instance vectors, which can affect subsequent tasks. ADMap consists of three main modules: Multi-scale Perception Neck (MPN), Instance Interactive Attention (IIA), and Vector Direction Difference Loss (VDDL). MPN captures multi-scale features to accurately construct instances of varying sizes. IIA facilitates interaction between instance and point-level information, enhancing the accuracy of point-level predictions. VDDL models the vector direction differences to refine the point sequence prediction process. The effectiveness of ADMap is demonstrated through extensive experiments on the nuScenes and Argoverse2 datasets, showing significant improvements in performance and stability compared to existing methods. ADMap achieves state-of-the-art results and is efficient in real-time map construction, making it a valuable tool for advancing HD map construction in autonomous driving.The paper introduces ADMap, an Anti-Disturbance Map construction framework designed to improve the quality of vectorized HD maps in autonomous driving. The framework addresses the issue of point sequence jitter and jaggedness in instance vectors, which can affect subsequent tasks. ADMap consists of three main modules: Multi-scale Perception Neck (MPN), Instance Interactive Attention (IIA), and Vector Direction Difference Loss (VDDL). MPN captures multi-scale features to accurately construct instances of varying sizes. IIA facilitates interaction between instance and point-level information, enhancing the accuracy of point-level predictions. VDDL models the vector direction differences to refine the point sequence prediction process. The effectiveness of ADMap is demonstrated through extensive experiments on the nuScenes and Argoverse2 datasets, showing significant improvements in performance and stability compared to existing methods. ADMap achieves state-of-the-art results and is efficient in real-time map construction, making it a valuable tool for advancing HD map construction in autonomous driving.