Real Time Local Wind Inference for Robust Autonomous Navigation

A new PhD thesis from the University of Pennsylvania presents a breakthrough in enabling aerial robots to navigate complex, windy urban environments. Researcher Spencer Folk developed a real-time system that fuses data from on-board navigational LiDAR with sparse, in-situ wind measurements to predict local flow fields. This deep learning model, drawing from fluid mechanics, allows drones to infer wind conditions without prior environmental knowledge. The research answers key questions about the sufficiency of topographical data for accurate prediction and the utility of learned models for planning.

The framework was integrated into a receding-horizon optimal controller to study its impact on autonomous flight. In simulated urban wind scenarios, the system demonstrated quantifiable improvements, reducing crash rates and lowering energy consumption during navigation. Crucially, sub-scale free flight experiments in an open-air wind tunnel proved these algorithms can run in real-time on an embedded flight computer, providing sufficient bandwidth for stable control of a small aerial robot. This establishes a new paradigm for robust motion planning in unknown, dynamic airspaces.