Reproduce “LO-World: Unified BEV Latent Modeling for Joint 4D Occupancy and LiDAR Forecasting” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhimin Liao, Ping Wei, Kai Mao, Shuaijia Chen

LO-World forecasts semantic occupancy and LiDAR together by aligning both modalities in a shared bird’s-eye-view latent, forecasting it once, and decoding paired outputs. On nuScenes, its trajectory-conditioned model reports 57.46% average IoU versus 51.80% for GenieDrive, while its trajectory-, box-, and map-conditioned variant reports 0.58 m² average Chamfer Distance versus 0.83 m² for LaGen. These comparisons use different input modalities.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.