LAformer: Trajectory Prediction for Autonomous Driving with Lane-Aware Scene Constraints

Abstract

Existing trajectory prediction methods for autonomous driving typically rely on one-stage trajectory prediction models, which condition future trajectories on observed trajectories combined with fused scene information. However, they often struggle with complex scene constraints, such as those encountered at intersections. To this end, we present a novel method, called LAformer. It uses an attention-based temporally dense lane-aware estimation module to continuously estimate the likelihood of the alignment between motion dynamics and scene information extracted from an HD map. Additionally, unlike one-stage prediction models, LAformer utilizes predictions from the first stage as anchor trajectories. It leverages a second-stage motion refinement module to further explore temporal consistency across the complete time horizon. Extensive experiments on nuScenes and Argoverse 1 demonstrate that LAformer achieves excellent generalized performance for multimodal trajectory prediction. The source code of LAformer is available at https://github.com/mengmengliu1998/LAformer.