DynFly: Dynamic-Aware Continuous Trajectory Generation for UAV Vision-Language Navigation in Urban Environments

2026-06-30

AI Summary

Existing vision-language navigation (VLN) methods for unmanned aerial vehicles (UAVs) typically produce discrete actions or sparse waypoints, struggling to generate continuous, stable, and executable trajectories. This work proposes DynFly, a novel framework that introduces, for the first time in UAV-VLN, a dynamics-aware continuous trajectory generation mechanism. DynFly employs a lightweight trajectory generation layer to map high-level navigation intents into B-spline parameterized trajectories, leveraging a Spline-DiT generator trained via flow matching. It further incorporates a multidimensional dynamic supervision signal encompassing position, velocity, acceleration, heading, and local goal alignment. The method seamlessly integrates into existing pipelines and achieves significant performance gains on the OpenUAV Test Unseen Full benchmark, improving NDTW, SDTW, SR, and OSR by 4.69, 2.40, 2.14, and 4.87 points, respectively, while reducing endpoint error by 4.51 meters.

Abstract

Recent advances in multimodal large models have significantly improved UAV vision-language navigation (UAV-VLN) by enhancing high-level perception and reasoning. However, existing methods mainly focus on predicting discrete actions, local targets, or sparse waypoints, while the continuous transition from navigation intent to executable UAV motion remains weakly modeled. This motion-interface gap limits the continuity, stability, and executability of generated UAV trajectories. To address this gap, we propose DynFly, a dynamic-aware continuous trajectory generation framework that bridges high-level navigation reasoning and executable UAV motion. DynFly bridges high-level navigation intent and continuous UAV motion through a lightweight trajectory generation layer. Specifically, it represents expert trajectories in B-spline control-point space and employs a Spline-DiT generator to learn conditional trajectory generation via flow matching. Furthermore, we introduce UAV-oriented dynamic-aware supervision over position, finite-difference velocity, finite-difference acceleration, heading consistency, and local target alignment, enabling the generated trajectories to better satisfy UAV motion characteristics. And our trajectory generation framework can also be integrated with an existing UAV-VLN framework while preserving its original visual-language reasoning pipeline. Extensive experiments on the OpenUAV UAV-VLN benchmark show that DynFly improves both navigation performance and trajectory quality. On the Test Unseen Full split, DynFly improves the strongest baseline by 4.69 NDTW, 2.40 SDTW, 2.14 SR points and 4.87 OSR points, while reducing NE by 4.51 m.

Problem

• UAV vision-language navigation
• continuous trajectory generation
• motion-interface gap
• trajectory executability
• dynamic-aware supervision

Innovation

• continuous trajectory generation
• dynamic-aware supervision
• B-spline control points
• flow matching
• UAV vision-language navigation