Real-time Avoidance of Weather Hazards In Dynamic Environments (RAWHIDE)

Unmanned systems play a critical military role, and require the ability to “fly through areas of operations on a near-all-weather 24/7 basis” so “potential weather-related incidents can be avoided” (OSD, 2013). To increase UAS flight safety and avoid weather phenomena (e.g., clouds, haze, ash), we propose a system for Real-time Avoidance of Weather Hazards In Dynamic Environments (RAWHIDE) to dynamically fuse electro-optical and infra-red sensor (EO/IR) data with onboard weather radar. RAWHIDE uses a FACE-compliant open system architecture to classify cloud regions with 2D segmentation based on a proven pipeline of region partitioning, feature extraction, classification, and tracking. This 2D cloud map is then fused with a 3D cloud model generated using a structure-from-motion algorithm. Aircraft weather radar images cue the EO/IR sensor to regions of high precipitation returns, independently validate the visual classifications, and provide accurate distance measurements. When clouds are detected, an A* path replanning algorithm generates an avoidance maneuver. Our in-house flight simulator generates synthetic cloud data and validates flight routes as part of the closed loop evaluation framework, and we will test on real and synthetic clouds across many operating conditions (e.g., solar angle, cloud type, wind, range).