DreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation
Teaching drones to follow spoken directions by remembering where they've been
Researchers developed DreamFly, a system that helps aerial drones navigate using spoken instructions by combining three key improvements: keeping track of what the drone has seen recently, planning multiple steps ahead but only executing one at a time, and explicitly deciding when the mission is complete. The system achieved 32% success rates on unseen environments—higher than all existing methods—while making fewer navigation errors overall.
Drones that reliably follow human instructions could speed up search-and-rescue operations, infrastructure inspections, and autonomous delivery. Current systems struggle with partial information and poor planning horizons, causing them to fail or overshoot targets. This approach tackles those specific problems through better memory and decision-making, directly improving success rates on real navigation tasks.