Abstract: Many existing end-to-end autonomous driving methods involve reinforcement learning or multi-stage discrete task pipelines. Reinforcement learning approaches lack deterministic ...