Real-Time GSplat Mapping

Neural radiance fields (NeRFs) and Gaussian Splats (GSplats) are a class of implicit scene representations that model 3D environments from color images. These scene representations are expressive, and can model the complex and multi-scale geometry of real world environments, which potentially makes them a powerful tool for robotics applications. Modern NeRF/GSplat training libraries can generate a photo-realistic volumetric scene representation from a static data set in just a few seconds, but are designed for offline use and require a slow pose optimization pre-computation step.Here we demonstrate SplatBridge, an open-source bridge between the Robot Operating System (ROS) and the popular Nerfstudio library for real-time, online training of NeRFs/GSplats from a stream of images. SplatBridge enables rapid development of research on applications of GSplats in robotics by providing an extensible interface to the efficient training pipelines and model libraries provided by Nerfstudio. In this demo, we show this tool applied to mapping and semantically identifying objects in a field environment.