DRIFT: Drift and Aggregation for Motion Planning

2026-07-15

AI Summary

This work addresses the critical challenge of efficiently generating executable driving trajectories under real-time constraints in autonomous driving. The authors propose DRIFT, a fixed-depth end-to-end trajectory planner that produces multimodal candidate trajectories via a single-step drift operation in a compact latent space. A novel scene-aware aggregation mechanism—requiring no trajectory-quality labels—directly outputs the final trajectory by fusing conditional features from a pretrained vision encoder, a lightweight aggregation head, map-boundary regularization, and expert imitation learning. Evaluated on the NAVSIM navtest benchmark, DRIFT achieves 89.6 PDMS and 90.4 EPDMS while running at 66.43 ms per inference on an RTX 4090, with all generated trajectories strictly adhering to drivable-area constraints.

Abstract

End-to-end trajectory planners need to represent multiple plausible driving behaviors while producing a single executable trajectory under real-time constraints. Proposal-based approaches address this ambiguity by generating multiple candidates, but converting the proposal set into a final plan remains a key design problem. We present DRIFT, a fixed-depth planner that combines one-step drifting in a compact trajectory latent space with scene-aware proposal aggregation. Conditioned on features from a pretrained visual encoder, the DRIFT Decoder generates 48 proposal features in a single batched pass, with 32 samples at alpha=0.5 and 16 samples at alpha=0.9. A lightweight Aggregation Head integrates these features with scene, navigation, and ego-state information and directly predicts the final trajectory without requiring trajectory-level quality labels for aggregation. Its output is trained with expert-trajectory imitation and a map-derived boundary regularizer that penalizes waypoints outside the drivable polygon and inside waypoints near its boundary. On NAVSIM navtest, DRIFT achieves 89.6 PDMS and 90.4 EPDMS, with strong drivable-area compliance and ego progress among the methods compared. The proposal-generation and aggregation module runs in 10.82 ms on an NVIDIA RTX 4090, while full-model inference including the visual backbone takes 66.43 ms. These results show that one-step latent proposal generation and direct aggregation provide an efficient design for multi-hypothesis motion planning.

Problem

• motion planning
• trajectory aggregation
• multi-hypothesis planning
• end-to-end planning
• proposal-based planning

Innovation

• one-step drifting
• latent trajectory space
• proposal aggregation
• drivable-area regularization
• end-to-end motion planning