End-to-end autonomous driving stack integrating perception, prediction, and planning with sparse scene representations for efficiency and safety.
