Existing data augmentation approaches on LiDAR 1 point cloud are mostly developed on rigid transformation, such 2 as rotation, flipping, or copy-based and mix-based methods, 3 lacking the capability to generate diverse samples that depict 4 smooth deformations in real-world scenarios. In response, we 5 propose a novel and effective LiDAR point cloud augmentation 6 approach with smooth deformations that can enrich the diversity 7 of training data while keeping the topology of instances and 8 scenes simultaneously. The whole augmentation pipeline can 9 be separated into two different parts: scene augmentation and 10 instance augmentation. To simplify the selection of deformation 11 functions and ensure control over augmentation outcomes, we 12 propose three effective strategies: residual mapping, space decou-13 pling, and function periodization, respectively. We also propose 14 an effective prior-based location sampling algorithm to paste 15 instances on a more reasonable area in the scenes. Extensive ex-16 periments on both the SemanticKITTI and nuScenes challenging 17 datasets demonstrate the effectiveness of our proposed approach 18 across various baselines. The codes are publicly available at 19 https://github.com/skyshoumeng/SmoothDA.20