World2Minecraft: Occupancy-Driven Simulated Scenes Construction

Researchers introduced World2Minecraft, a system that reconstructs real indoor scenes as editable Minecraft worlds using semantic occupancy predictions.

The project also released MinecraftOcc, a 100,165-image dataset spanning 156 indoor scenes to benchmark and improve occupancy models.

The approach offers a flexible simulation platform for embodied AI, including navigation and Vision-Language Navigation tasks.

By turning real environments into structured Minecraft scenes, the work aims to support stronger 3D scene understanding and training data for robotics research.