Recently, Cainiao has developed the industry's first high-performance large-scale reinforcement learning planner for autonomous driving, which will enhance Cainiao's unmanned vehicles' ability to cope with complex scenarios. The research results have been included in the top international academic conference CVPR 2025. CVPR 2025, the top international conference in the field of computer vision and pattern recognition hosted by IEEE, recently announced the results of the joint research submitted by Zhejiang University and Cainiao's unmanned vehicle technical team. The conference has accepted the results. The joint research team proposed a new consistent autoregressive planner CarPlanner (Consistent Auto-regressive Trajectory Planning for Large-scale Reinforcement Learning in Autonomous Driving), which solves the problems of inefficiency and insufficient performance of reinforcement learning in trajectory planning. On the extremely demanding large-scale real-world dataset nuPlan, Cainiao proved for the first time that a planner based on reinforcement learning can surpass solutions based on imitation learning and rules. This is also the industry's first high-performance large-scale reinforcement learning planner for autonomous driving.