This significant upgrade of HSD focuses on a leading architecture of "one-stage end-to-end + reinforcement learning." Unlike ...