The paper introduces CG-SLAM, an efficient dense RGB-D SLAM system based on a novel uncertainty-aware 3D Gaussian field. This system aims to address the computational intensity and time consumption issues of existing NeRF-based methods, which are hindered by their volume rendering pipelines. By analyzing Gaussian Splatting, the authors propose techniques to construct a consistent and stable 3D Gaussian field suitable for tracking and mapping. They also introduce a depth uncertainty model to guide the selection of valuable Gaussian primitives during optimization, improving tracking efficiency and accuracy. Experiments on various datasets demonstrate that CG-SLAM achieves superior performance in tracking and mapping, with a notable tracking speed of up to 15 Hz. The system's effectiveness is validated through extensive evaluations, showing its superior accuracy, efficiency, and robustness compared to state-of-the-art methods.The paper introduces CG-SLAM, an efficient dense RGB-D SLAM system based on a novel uncertainty-aware 3D Gaussian field. This system aims to address the computational intensity and time consumption issues of existing NeRF-based methods, which are hindered by their volume rendering pipelines. By analyzing Gaussian Splatting, the authors propose techniques to construct a consistent and stable 3D Gaussian field suitable for tracking and mapping. They also introduce a depth uncertainty model to guide the selection of valuable Gaussian primitives during optimization, improving tracking efficiency and accuracy. Experiments on various datasets demonstrate that CG-SLAM achieves superior performance in tracking and mapping, with a notable tracking speed of up to 15 Hz. The system's effectiveness is validated through extensive evaluations, showing its superior accuracy, efficiency, and robustness compared to state-of-the-art methods.