Problem
Research questions and friction points this paper is trying to address.
Long-Horizon Planning
End-to-End Autonomous Driving
Historical Reliability
Motion-Stage Evolution
Not All History Helps: Velocity-Aware Selective Memory for Long-Horizon End-to-End Autonomous Driving
🤖 AI Summary
This study addresses the challenges of unreliable historical states and motion evolution in long-horizon planning for end-to-end autonomous driving by proposing StableDrive. The method leverages Mamba operators to construct a selective momentum memory that enhances the robustness of historical representations, while introducing a motion-stage training scaffold to guide the model in perceiving dynamic evolution, thereby enabling efficient single-model deployment without ensembling. Experiments demonstrate that StableDrive achieves state-of-the-art performance on benchmarks such as nuScenes, reducing collision rates by 23.3% and attaining the highest EPDMS score on NAVSIM v2. These results indicate significant improvements in both safety and temporal consistency for long-horizon planning, validating the effectiveness of integrating structured memory mechanisms with stage-aware training in complex driving scenarios.
📝 Abstract
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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Selective Momentum Memory
Mamba State-Space
Motion-Stage Training Scaffold
Long-Horizon Planning
End-to-End Autonomous Driving
🔎 Similar Papers
No similar papers found.
