Loopy-SLAM is a dense neural SLAM system that improves upon existing methods by incorporating loop closures to achieve globally consistent maps and accurate trajectory estimation. The system uses neural point cloud submaps that grow iteratively during scene exploration, allowing for efficient and accurate mapping and tracking. Unlike previous methods that require storing the entire history of input frames, Loopy-SLAM anchors neural features in point cloud submaps that can be shifted without recomputing the dense map from scratch. This approach enables efficient map corrections and avoids the need for expensive gradient-based updates. The system uses a data-driven point-based submap generation method and triggers loop closures online through global place recognition. Robust pose graph optimization is used to rigidly align local submaps, and the system is evaluated on synthetic and real-world datasets, showing competitive or superior performance in tracking, mapping, and rendering accuracy compared to existing dense neural RGBD SLAM methods. The system is particularly effective in reducing drift and improving the accuracy of reconstructions, especially in complex scenes. Loopy-SLAM is implemented using a neural point cloud representation, which allows for efficient and accurate mapping and tracking, and is evaluated on the Replica, TUM-RGBD, and ScanNet datasets. The results show that Loopy-SLAM achieves state-of-the-art performance in terms of geometric reconstruction accuracy, tracking, and rendering. The system is also efficient in terms of memory and runtime, and is capable of running on a 12 GB GPU. The system is designed to be robust and scalable, and is able to handle large-scale scenes with minimal drift. The system is also able to handle complex scenes with high accuracy, and is able to outperform existing methods in terms of both tracking and rendering accuracy. The system is also able to handle real-time data, and is able to perform online loop closure to improve the accuracy of reconstructions. The system is able to handle a wide range of scenes, and is able to achieve high accuracy in both tracking and rendering. The system is also able to handle complex scenes with high accuracy, and is able to outperform existing methods in terms of both tracking and rendering accuracy. The system is also able to handle real-time data, and is able to perform online loop closure to improve the accuracy of reconstructions. The system is able to handle a wide range of scenes, and is able to achieve high accuracy in both tracking and rendering.Loopy-SLAM is a dense neural SLAM system that improves upon existing methods by incorporating loop closures to achieve globally consistent maps and accurate trajectory estimation. The system uses neural point cloud submaps that grow iteratively during scene exploration, allowing for efficient and accurate mapping and tracking. Unlike previous methods that require storing the entire history of input frames, Loopy-SLAM anchors neural features in point cloud submaps that can be shifted without recomputing the dense map from scratch. This approach enables efficient map corrections and avoids the need for expensive gradient-based updates. The system uses a data-driven point-based submap generation method and triggers loop closures online through global place recognition. Robust pose graph optimization is used to rigidly align local submaps, and the system is evaluated on synthetic and real-world datasets, showing competitive or superior performance in tracking, mapping, and rendering accuracy compared to existing dense neural RGBD SLAM methods. The system is particularly effective in reducing drift and improving the accuracy of reconstructions, especially in complex scenes. Loopy-SLAM is implemented using a neural point cloud representation, which allows for efficient and accurate mapping and tracking, and is evaluated on the Replica, TUM-RGBD, and ScanNet datasets. The results show that Loopy-SLAM achieves state-of-the-art performance in terms of geometric reconstruction accuracy, tracking, and rendering. The system is also efficient in terms of memory and runtime, and is capable of running on a 12 GB GPU. The system is designed to be robust and scalable, and is able to handle large-scale scenes with minimal drift. The system is also able to handle complex scenes with high accuracy, and is able to outperform existing methods in terms of both tracking and rendering accuracy. The system is also able to handle real-time data, and is able to perform online loop closure to improve the accuracy of reconstructions. The system is able to handle a wide range of scenes, and is able to achieve high accuracy in both tracking and rendering. The system is also able to handle complex scenes with high accuracy, and is able to outperform existing methods in terms of both tracking and rendering accuracy. The system is also able to handle real-time data, and is able to perform online loop closure to improve the accuracy of reconstructions. The system is able to handle a wide range of scenes, and is able to achieve high accuracy in both tracking and rendering.