SkyAID: A Safe Autonomy Infrastructure for Disaster Response

Phase II builds on our Phase I accomplishments by advancing the SkyAID system into a fully integrated prototype capable of addressing NASA’s wildfire response challenges in complex, partially known environments. We will transition simulation-validated SkyAID functionalities into a tangible hardware platform that combines GPS, IMU, and cameras for autonomous navigation under diverse lighting and visibility conditions. This development will encompass mechanical design, the creation of a digital twin, implementation and calibration of the perception system, and comprehensive field experiments to validate overall performance. Building on the initial development of AerDETR—a deep learning-based object detection system for aerial platforms—we will enhance system robustness in wildfire-prone scenarios by generating high-fidelity simulated aerial datasets with labeled objects to train perception networks capable of operating under adverse conditions such as smoke, variable lighting, and seasonal changes. Concurrently, we will develop a sensor fusion framework that integrates data from LIDAR, infrared, and optical cameras. We will perform rigorous verification and validation of the SkyAID control system. We will also develop specialized control strategies for distributed sensing of wind and fire conditions using swarms of drones by simulating a distributed sensing architecture where multiple drones collect environmental data to generate real-time fire and wind maps, and by implementing cooperative decision-making algorithms for adaptive waypoint planning and robust swarm coordination, with hardware experiments validating multi-drone fire sensing and communication. Collectively, these efforts will deliver a prototype hardware system and a comprehensive software toolbox that not only demonstrate reliable sim-to-real transfer but also provide scalable, multi-modal autonomy for effective wildfire reconnaissance, mapping, and suppression.