发表机构
Nanyang Technological University; North University of China; Yanshan University; Beijing Jiaotong University; China Automotive Technology and Research Center Co., Ltd.; Harbin Institute of Technology; Beijing Institute of Technology（南洋理工大学; 中北大学; 燕山大学; 北京交通大学; 中国汽车技术研究中心有限公司; 哈尔滨工业大学; 北京理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对长 horizon 端到端自动驾驶的历史规划状态不可靠问题，提出 StableDrive 模型，通过 SMM 与 MSTS 实现规划优化，在多个数据集上取得规划指标的 SOTA 性能。
AI 中文摘要
可靠的长 horizon 规划仍是端到端自动驾驶的关键挑战。通过考量未来运动演化与潜在后果，它为动态交通环境中安全、一致的驾驶提供前瞻性指导。现有方法将历史规划状态作为时间上下文，但自生成的历史可能过时或与当前运动阶段冲突，引入不可靠先验。我们提出 StableDrive 以解决跨周期历史可靠性与 horizon 内运动阶段演化问题。选择性动量记忆（Selective Momentum Memory, SMM）采用 Mamba 选择性状态空间算子实现，用于控制前序自预测规划状态对当前周期的影响。运动阶段训练支架（Motion-Stage Training Scaffold, MSTS）利用运动阶段、长 horizon 轨迹及纵向运动监督，指导阶段感知的未来运动学习，推理前移除该支架。架构对齐的两个端点间的固定参数中点生成单个可部署的 SMM 规划器，无需模型集成或额外推理时间计算。在 nuScenes 数据集的 MomAD 评估协议下，StableDrive 在 1 至 6 秒的所有报告规划指标上达到 SOTA 性能，较各指标之前的最优值，平均碰撞率降低 23.3%，TPC 降低 30.9%，L2 降低 11.8%；在经整理的纵向过渡 nuScenes（LT-nuScenes）数据集上，StableDrive 将 6 秒碰撞率降低 23.81%，TPC 降低 10.90%，L2 降低 6.37%；在 NAVSIM v1 和 v2 上，StableDrive 在所有三个报告设置中取得最高 PDMS/EPDMS，其中 v2 navhard 上的 EPDMS 较之前最优值提升 5.7 个点。
英文摘要
Reliable long-horizon planning remains a key challenge in end-to-end autonomous driving. By accounting for future motion evolution and potential consequences, it provides forward-looking guidance for safe and consistent driving in evolving traffic environments. Existing methods use historical planning states as temporal context. Self-generated history may become stale or conflict with the current motion stage, introducing unreliable priors. We propose StableDrive to address cross-cycle historical reliability and within-horizon motion-stage evolution. Selective Momentum Memory (SMM), implemented with a Mamba selective state-space operator, controls the influence of the preceding self-predicted planning state on the current cycle. Motion-Stage Training Scaffold (MSTS) uses motion-stage, long-horizon trajectory, and longitudinal-motion supervision to guide stage-aware future motion learning and is removed before inference. A fixed parameter midpoint between two architecture-aligned endpoints yields a single deployable SMM planner without model ensembling or extra inference-time computation. On nuScenes under the MomAD evaluation protocol, StableDrive achieves SOTA performance across all reported planning metrics from 1 to 6 s, reducing average collision rate by 23.3%, TPC by 30.9%, and L2 by 11.8% over the best previously reported value for each metric. On the curated Longitudinal-Transition nuScenes (LT-nuScenes), StableDrive reduces 6-s collision rate by 23.81%, TPC by 10.90%, and L2 by 6.37%. On NAVSIM v1 and v2, StableDrive achieves the highest PDMS/EPDMS in all three reported settings, including a 5.7-point EPDMS gain on v2 navhard over the previous best.
Comments14 pages, 7 figures
