LiDAR4D: Dynamic Neural Fields for Novel Space-time View LiDAR Synthesis This paper proposes LiDAR4D, a differentiable LiDAR-only framework for novel space-time LiDAR view synthesis. The method addresses the challenges of dynamic scene reconstruction for LiDAR point clouds, including sparsity, large-scale reconstruction, and temporal consistency. LiDAR4D introduces a 4D hybrid representation combining multi-planar and grid features to achieve effective reconstruction in a coarse-to-fine manner. It also incorporates geometric constraints derived from point clouds to improve temporal consistency and global optimization of ray-drop probability to preserve cross-region patterns. Extensive experiments on KITTI-360 and NuScenes datasets demonstrate the superiority of LiDAR4D in accomplishing geometry-aware and time-consistent dynamic reconstruction. The method outperforms previous state-of-the-art NeRF-based implicit approaches and explicit reconstruction methods, achieving significant reductions in CD error on both datasets. LiDAR4D is capable of achieving geometry-aware and time-consistent reconstruction under large-scale dynamic scenarios. The main contributions include proposing LiDAR4D, introducing 4D hybrid neural representations and motion priors derived from point clouds for geometry-aware and time-consistent large-scale scene reconstruction, and demonstrating the state-of-the-art performance of LiDAR4D in challenging dynamic scene reconstruction and novel view synthesis. The method is evaluated on diverse dynamic scenarios of KITTI-360 and NuScenes autonomous driving datasets, showing significant improvements in range depth and intensity metrics. The paper also discusses related work, including LiDAR simulation, neural radiance fields, and dynamic scene reconstruction. The methodology includes problem formulation, preliminary for NeRF, 4D hybrid planar-grid representation, scene flow prior, neural LiDAR fields, ray-drop refinement, and optimization. The experiments demonstrate the effectiveness of LiDAR4D in novel-view LiDAR synthesis, with quantitative comparisons on KITTI-360 and NuScenes datasets. The method is evaluated on both static and dynamic scenes, showing superior performance in depth and intensity reconstruction. The paper also discusses limitations, including long-distance vehicle motion and occlusion problems, and concludes that LiDAR4D is a novel framework for dynamic scene reconstruction and synthesis.LiDAR4D: Dynamic Neural Fields for Novel Space-time View LiDAR Synthesis This paper proposes LiDAR4D, a differentiable LiDAR-only framework for novel space-time LiDAR view synthesis. The method addresses the challenges of dynamic scene reconstruction for LiDAR point clouds, including sparsity, large-scale reconstruction, and temporal consistency. LiDAR4D introduces a 4D hybrid representation combining multi-planar and grid features to achieve effective reconstruction in a coarse-to-fine manner. It also incorporates geometric constraints derived from point clouds to improve temporal consistency and global optimization of ray-drop probability to preserve cross-region patterns. Extensive experiments on KITTI-360 and NuScenes datasets demonstrate the superiority of LiDAR4D in accomplishing geometry-aware and time-consistent dynamic reconstruction. The method outperforms previous state-of-the-art NeRF-based implicit approaches and explicit reconstruction methods, achieving significant reductions in CD error on both datasets. LiDAR4D is capable of achieving geometry-aware and time-consistent reconstruction under large-scale dynamic scenarios. The main contributions include proposing LiDAR4D, introducing 4D hybrid neural representations and motion priors derived from point clouds for geometry-aware and time-consistent large-scale scene reconstruction, and demonstrating the state-of-the-art performance of LiDAR4D in challenging dynamic scene reconstruction and novel view synthesis. The method is evaluated on diverse dynamic scenarios of KITTI-360 and NuScenes autonomous driving datasets, showing significant improvements in range depth and intensity metrics. The paper also discusses related work, including LiDAR simulation, neural radiance fields, and dynamic scene reconstruction. The methodology includes problem formulation, preliminary for NeRF, 4D hybrid planar-grid representation, scene flow prior, neural LiDAR fields, ray-drop refinement, and optimization. The experiments demonstrate the effectiveness of LiDAR4D in novel-view LiDAR synthesis, with quantitative comparisons on KITTI-360 and NuScenes datasets. The method is evaluated on both static and dynamic scenes, showing superior performance in depth and intensity reconstruction. The paper also discusses limitations, including long-distance vehicle motion and occlusion problems, and concludes that LiDAR4D is a novel framework for dynamic scene reconstruction and synthesis.