See-and-Reach: Precise Vision-Language Navigation for UAVs within the Field of View
Researchers have introduced UAV-VLN-FOV, a new task designed to evaluate the precise 'see-and-reach' capability of Unmanned Aerial Vehicles (UAVs) when a target is within their field of view. This task isolates the final approach phase of navigation, distinct from longer-range target discovery. To enhance this capability, they also proposed 3DG-VLN, a framework that uses dynamic 3D direction cues and high-resolution observations to improve visual grounding and spatial alignment for accurate target reaching. Experiments demonstrated that 3DG-VLN achieved a 13.82% improvement in success rate over existing baselines, with real-world trials showing its practical potential. AI
IMPACT This research could lead to more precise autonomous navigation systems for drones in complex environments.