Search papers, labs, and topics across Lattice.
This paper introduces MomADv2, a state-space memory framework designed to enhance long-horizon planning in autonomous driving by selectively filtering historical planning queries based on temporal continuity and command consistency. The framework addresses the challenge of invalid memory that can mislead decision-making when driving commands change, thus improving planning reliability. Experimental results show that MomADv2 achieves a 15.6% reduction in average collision rates compared to its predecessor, MomAD, while maintaining planning stability over a 6-second horizon.
Command-inconsistent memory can lead to a 15.6% increase in collision rates during autonomous driving, but MomADv2 effectively filters this noise for safer long-horizon planning.
Long-horizon planning is critical for safe autonomous driving in complex scenarios. Existing methods improve planning continuity with temporal memory, but such memory may become invalid and mislead decisions when the driving command changes. Thus, selectively leveraging useful history while suppressing command-inconsistent memory remains a key challenge. To address this issue, we propose MomADv2, a reliable state-space memory framework for long-horizon end-to-end autonomous driving. At its core, MomADv2 introduces a Selective State-Space Planning Memory Query Module, which filters historical planning queries based on temporal continuity and command consistency, selects planning modes relevant to the current command, and models the evolution of planning intentions through a selective state-space mechanism. To further alleviate local trajectory deviations and error accumulation in long-horizon planning, we design a Flow-Matching Trajectory Residual Refiner. It learns a continuous residual correction field from the refined planning output to the expert trajectory, enabling fine-grained trajectory refinement while preserving the stability of anchor-based planning. Extensive experiments on closed-loop NAVSIM and Bench2Drive, as well as open-loop nuScenes, demonstrate that MomADv2 improves long-horizon planning consistency and reduces the average collision rate by 15.6% over MomAD under 6-second planning.
Tsinghua AI