FASTopoWM: Fast-Slow Lane Segment Topology Reasoning with Latent World Models

ICLR 2026 Conference Submission351 Authors

Keywords: Lane Topology Reasoning, Fast-slow System, World Model, Autonomous Driving

Abstract: Lane segment topology reasoning provides comprehensive bird's-eye view (BEV) road scene understanding, which can serve as a key perception module in planning-oriented end-to-end autonomous driving systems. Current approaches prioritize graph modeling, endpoint alignment, and multi-attribute learning, yet they often neglect temporal modeling. This leads to inconsistent inter-frame detection within scene flows and motivates our focus on temporal propagation for lane segments. Recently, stream-based methods have shown promising outcomes by integrating temporal cues at both the query and BEV levels. However, it remains limited by over-reliance on historical queries, vulnerability to pose estimation failures, and insufficient temporal propagation. To overcome these limitations, we propose FASTopoWM, a novel fast-slow lane segment topology reasoning framework augmented with latent world models. To reduce the impact of pose estimation failures, this unified framework enables parallel supervision of both historical and newly initialized queries, facilitating mutual reinforcement between the fast and slow systems. Furthermore, we introduce latent query and BEV world models conditioned on the action latent to propagate the state representations from past observations to the current timestep. This design substantially improves the performance of temporal perception within the slow pipeline. Extensive experiments on the OpenLane-V2 benchmark demonstrate that FASTopoWM outperforms state-of-the-art methods in both lane segment detection and centerline perception. Our code will be released.

Primary Area: applications to robotics, autonomy, planning

Submission Number: 351