The paper "S³Gaussian: Self-Supervised Street Gaussians for Autonomous Driving" introduces a novel method for 3D scene reconstruction in autonomous driving scenarios. The authors address the challenge of decomposing dynamic and static elements in street scenes without requiring explicit 3D annotations, which is a common limitation of existing methods. The proposed method, S³Gaussian, uses 3D Gaussian Splatting (3DGS) to represent scenes explicitly and efficiently, along with a spatial-temporal field network to model 4D dynamics. This network includes a multi-resolution Hexplane structure encoder and a multi-head Gaussian decoder, which together capture and decompose the scene into static and dynamic components. Extensive experiments on the Waymo-Open dataset demonstrate that S³Gaussian outperforms state-of-the-art methods in both scene reconstruction and novel view synthesis, achieving high-quality rendering and accurate dynamic object representation. The method's effectiveness is further validated through qualitative and quantitative evaluations, showing superior performance in handling complex and dynamic urban scenes.The paper "S³Gaussian: Self-Supervised Street Gaussians for Autonomous Driving" introduces a novel method for 3D scene reconstruction in autonomous driving scenarios. The authors address the challenge of decomposing dynamic and static elements in street scenes without requiring explicit 3D annotations, which is a common limitation of existing methods. The proposed method, S³Gaussian, uses 3D Gaussian Splatting (3DGS) to represent scenes explicitly and efficiently, along with a spatial-temporal field network to model 4D dynamics. This network includes a multi-resolution Hexplane structure encoder and a multi-head Gaussian decoder, which together capture and decompose the scene into static and dynamic components. Extensive experiments on the Waymo-Open dataset demonstrate that S³Gaussian outperforms state-of-the-art methods in both scene reconstruction and novel view synthesis, achieving high-quality rendering and accurate dynamic object representation. The method's effectiveness is further validated through qualitative and quantitative evaluations, showing superior performance in handling complex and dynamic urban scenes.