SatNav: Scalable Long-Horizon UAV Vision-Language Navigation Benchmark From Satellite Imagery
Jiajun Jiang · hf · 2026-10-09
Researchers introduce SatNav, a scalable long-horizon UAV vision-language navigation benchmark built from high-resolution satellite imagery, avoiding costly 3D reconstruction.
- 118K episodes from 59 scenes across 18 cities, average trajectory 379 m, targeting city-scale navigation
- Three task families (Boundary, Landmark, Route) stress-test long-horizon memory and geospatial grounding
- Benchmarking shows city-scale navigation remains hard for classic VLN agents and LVLM-based agents
- Proposes SwiftVLN, a modular framework with switchable memory components, plus systematic memory ablations
- Satellite-to-UAV transfer experiments show satellite-trained models work on real-flight observations
More from Embodied
- CARE certifies VLA inference speedups up to 10.8x with statistical guarantees — UMCP · 2026-10-09
- Corporate robot bets graded: only one binding order with unit counts found — lukas_m_ziegler · 2026-10-09
- Hot-reload Lua on a virtual ESP32 and vibe-code hardware live from Claude — genmon · 2026-10-09
- New LeRobot dataset visualizer could triple robotics dev speed, says team member — k7agar · 2026-10-09
- RoboRender Turns Simulated Trajectories Into Photorealistic Video, Hits 71% Zero-Shot Real-Robot Success — YouJiacheng · 2026-10-09
- Open-Source SOTA Robotics Model Goes Head-to-Head with Closed-Source General Model — eigenron · 2026-10-09