Threat-Aware UAV Dodging of Human-Thrown Projectiles with an RGB-D Camera

HE rapid advancement of uncrewed aerial vehicles (UA Vs) and their supporting infrastructure has significantly expanded the UA V market, enabling diverse applications such as aerial imaging, last-mile delivery, and air traffic management [1], [2]. To meet the demands of these complex tasks, modern UA Vs are increasingly equipped with autonomous modules for environmental perception, navigation, and obstacle avoidance. Despite these advances, UA Vs often fail to cope with sudden human-initiated attacks. Recent reports have documented cases where crowds at public events throw projectiles to disrupt UA V operations [3], [4], posing significant threats to their safety and public security. Consequently, there is an urgent need for robust strategies to counter human-initiated attacks involving fast-moving projectiles. Developing robust UA V systems capable of rapid responses to sudden human-initiated attacks remains a critical and unresolved research problem. Dodging such projectile threats involves overcoming several challenges: (1) Perception Latency: Projectiles often emerge suddenly at close range, leaving a narrow time window for detection and dodging. Therefore, minimizing the delay between sensing and control is crucial while maintaining high prediction accuracy to ensure effective avoidance.