YUKUIv2: A Key-point Driven Diffusion Planner for Multi-shot Autonomous Parking

Multi-shot trajectory planning remains a critical challenge in autonomous driving and robotics, particularly when handling discontinuous motion states. To address this, we propose YUKUIv2, a novel Y-shaped trajectory planning module. Our approach introduces a key-point driven diffusion mechanism, explicitly predicting discrete transition points to guide the continuous generation of multi-shot trajectories. Evaluated on an CARLA parking dataset, the proposed planner effectively bridges motion state awareness with trajectory diffusion. Results demonstrate that YUKUIv2 maintains high planning accuracy while significantly improving generalization capabilities across complex environmental configurations.