ADMap is an anti-disturbance framework for vectorized HD map construction. The framework consists of three modules: Multi-scale Perception Neck (MPN), Instance Interactive Attention (IIA), and Vector Direction Difference Loss (VDDL). MPN captures multi-scale features in the BEV map, improving the accuracy of constructing instances with significant size differences. IIA flexibly encodes instance-level and point-level information, enabling the network to capture relationships between point-level embeddings. VDDL models the association between instance points and vector direction differences, using vector direction differences as losses to constrain the construction process of point sequences more precisely. ADMap achieves state-of-the-art performance on the nuScenes and Argoverse2 datasets. Extensive results demonstrate its ability to produce stable and reliable map elements in complex and changing driving scenarios. ADMap outperforms existing vectorized HD map models in both nuScenes and Argoverse2 benchmarks. In nuScenes, ADMap improves performance by 4.2% and 5.5% in camera-only and multimodal frameworks, respectively, compared to the baseline method MapTR. ADMapv2 not only reduces inference latency but also improves performance of the baseline method MapTRv2. ADMap also performs well in Argoverse2. ADMapv2 improves mAP by 62.9% while maintaining FPS at 14.8, demonstrating that ADMap is an efficient and high-precision framework for generating accurate and smooth map topology in complex scenes. The contributions of this paper are summarized as follows: (1) End-to-end framework ADMap is proposed, which constructs more stable vectorized HD maps. (2) MPN captures multi-scale information more precisely without increasing computational resources, IIA achieves effective interaction between inter-instance and intra-instance information to alleviate the problem of instance point position offset. VDDL models the vector direction difference and supervises the construction process of point order position using topological information. ADMap enables real-time construction of vectorized HD maps and achieves state-of-the-art in both the nuScenes and Argoverse2 benchmarks.ADMap is an anti-disturbance framework for vectorized HD map construction. The framework consists of three modules: Multi-scale Perception Neck (MPN), Instance Interactive Attention (IIA), and Vector Direction Difference Loss (VDDL). MPN captures multi-scale features in the BEV map, improving the accuracy of constructing instances with significant size differences. IIA flexibly encodes instance-level and point-level information, enabling the network to capture relationships between point-level embeddings. VDDL models the association between instance points and vector direction differences, using vector direction differences as losses to constrain the construction process of point sequences more precisely. ADMap achieves state-of-the-art performance on the nuScenes and Argoverse2 datasets. Extensive results demonstrate its ability to produce stable and reliable map elements in complex and changing driving scenarios. ADMap outperforms existing vectorized HD map models in both nuScenes and Argoverse2 benchmarks. In nuScenes, ADMap improves performance by 4.2% and 5.5% in camera-only and multimodal frameworks, respectively, compared to the baseline method MapTR. ADMapv2 not only reduces inference latency but also improves performance of the baseline method MapTRv2. ADMap also performs well in Argoverse2. ADMapv2 improves mAP by 62.9% while maintaining FPS at 14.8, demonstrating that ADMap is an efficient and high-precision framework for generating accurate and smooth map topology in complex scenes. The contributions of this paper are summarized as follows: (1) End-to-end framework ADMap is proposed, which constructs more stable vectorized HD maps. (2) MPN captures multi-scale information more precisely without increasing computational resources, IIA achieves effective interaction between inter-instance and intra-instance information to alleviate the problem of instance point position offset. VDDL models the vector direction difference and supervises the construction process of point order position using topological information. ADMap enables real-time construction of vectorized HD maps and achieves state-of-the-art in both the nuScenes and Argoverse2 benchmarks.