ActiveSplat is a new technology that uses Gaussian Splatting to create realistic 3D models of environments. It works by taking images of a space and converting them into a 3D representation using machine learning. This allows ActiveSplat to create detailed 3D models without requiring the creation of complex meshes. The technology uses a hybrid map representation that combines dense environmental information with a sparse abstraction of the space, enabling efficient exploration and reconstruction. The researchers behind ActiveSplat have designed a hierarchical planning strategy to ensure high-fidelity reconstruction and photorealistic view synthesis.