LO-World for Unified 4D Occupancy and LiDAR Forecasting
November 8, 2026
LO-World is a 4D world model that forecasts LiDAR point clouds and 3D semantic occupancy within a shared BEV latent space. The single-stream formulation uses a hybrid range-voxel autoencoder and flow-matching generator to allow metric geometry and semantics to reinforce each other.
HOW THIS AFFECTS YOU
●
researcherYou can use this shared latent space to improve how autonomous systems forecast both geometry and semantics simultaneously.