Researchers have developed World2Minecraft, a system that converts real-world scenes into structured Minecraft environments by predicting 3D semantic occupancy. This allows for downstream tasks like Vision-Language Navigation within these reconstructed scenes. To improve reconstruction quality, a new data acquisition pipeline and the large-scale MinecraftOcc dataset were created, featuring over 100,000 images from 156 indoor scenes. Additionally, a separate project, Dream-Cubed, introduces a large dataset of Minecraft worlds and models trained on billions of cubes for efficient, controllable generation of interactive 3D environments. AI
Summary written by gemini-2.5-flash-lite from 4 sources. How we write summaries →
IMPACT Enables more realistic and customizable simulation environments for embodied AI research and 3D generative modeling.
RANK_REASON The cluster contains two arXiv papers detailing new methods and datasets for generating and reconstructing 3D environments within Minecraft.