CG-SLAM is an efficient dense RGB-D SLAM system that uses a novel uncertainty-aware 3D Gaussian field for consistent and stable 3D reconstruction. The system improves tracking efficiency and accuracy by incorporating a depth uncertainty model to select valuable Gaussian primitives during optimization. Through an in-depth analysis of Gaussian splatting, CG-SLAM constructs a consistent and stable 3D Gaussian field suitable for tracking and mapping. The system achieves superior tracking and mapping performance with a tracking speed of up to 15 Hz on various datasets. Experiments show that CG-SLAM outperforms existing methods in terms of tracking accuracy, reconstruction quality, and runtime efficiency. The system uses a GPU-accelerated rasterizer to enable fast rendering and efficient tracking. The 3D Gaussian field is designed to be robust against overfitting and to provide accurate geometric information. The system also incorporates an uncertainty model to filter out unreliable Gaussian primitives and improve tracking accuracy. CG-SLAM is evaluated on multiple RGB-D datasets, demonstrating its effectiveness in real-world scenarios. The system's contributions include a new GPU-accelerated framework for real-time dense RGB-D SLAM, a consistent and stable 3D Gaussian field, and a novel depth uncertainty model. The system is expected to achieve better performance in tracking, mapping, and rendering compared to existing methods.CG-SLAM is an efficient dense RGB-D SLAM system that uses a novel uncertainty-aware 3D Gaussian field for consistent and stable 3D reconstruction. The system improves tracking efficiency and accuracy by incorporating a depth uncertainty model to select valuable Gaussian primitives during optimization. Through an in-depth analysis of Gaussian splatting, CG-SLAM constructs a consistent and stable 3D Gaussian field suitable for tracking and mapping. The system achieves superior tracking and mapping performance with a tracking speed of up to 15 Hz on various datasets. Experiments show that CG-SLAM outperforms existing methods in terms of tracking accuracy, reconstruction quality, and runtime efficiency. The system uses a GPU-accelerated rasterizer to enable fast rendering and efficient tracking. The 3D Gaussian field is designed to be robust against overfitting and to provide accurate geometric information. The system also incorporates an uncertainty model to filter out unreliable Gaussian primitives and improve tracking accuracy. CG-SLAM is evaluated on multiple RGB-D datasets, demonstrating its effectiveness in real-world scenarios. The system's contributions include a new GPU-accelerated framework for real-time dense RGB-D SLAM, a consistent and stable 3D Gaussian field, and a novel depth uncertainty model. The system is expected to achieve better performance in tracking, mapping, and rendering compared to existing methods.