無人機系統飛航監管之避撞策略與模擬

Abstract

本研究發展一套無人機之避撞策略及模擬驗證平臺。隨著無人機的普及，且操作門檻不高，加上缺乏完善的飛航監管機制下，無人機事故頻繁發生，使得民眾對其產生安全疑慮。為此開發無人機飛航監管系統之避撞策略，並參考無人機飛航監管相關規範及避撞相關研究。由於人為飛航監管成本高，且易發生誤判，因此本研究採用自主且即時的監管方式。研究中設有四種監管機制：機體追蹤、進場順序確認與更新、數量管控、飛航避撞及指示。飛航避撞為運算核心，利用當前接收到的飛航數據，以及建築物、空域圖資，並建置作為避撞緩衝的機體防護長方體、建築物及空域防護體。採用網格模型與查表法，快速地找出具飛航風險之目標，目標包含無人機、建築物及空域，並即時給予操作人適當的避撞指示。本研究利用AirSim模擬軟體建置模擬驗證平臺。模擬結果顯示能有效地協調多臺無人機的飛航與避撞，成功地避免飛安風險之發生。最大運算時間可於1秒之取樣時間內完成，此成果可作為未來無人機飛航監管系統建置之參考。後續可引入深度學習及分散式監管架構等技術，建置出具備持續更新、高效能及低系統容錯率的飛航監管之避撞策略。

In this study, collision avoidance strategies and a simulated verification platform for UAVs were developed. With the popularity of UAVs and the simpler operation, coupled with the lack of a comprehensive flight monitoring mechanism and system, UAV accidents have occurred frequently, causing safety issues for the public. To develop collision avoidance strategies, relevant regulations of flight monitoring mechanisms and collision avoidance researches were referred to in this study. Due to the high cost and misjudgment of manual flight monitoring, an authe tonomous and real-time control method was adopted. There are four control mechanisms: airframe tracking, approaching sequence confirmation and updating, quantity control, and flight collision avoidance and indication. Flight collision avoidance is the core of the whole process. It utilizes the currently received flight data, buildings, and airspace maps. During the flight collision avoidance step, protective rectangles around the airframe, buildings, and airspace were constructed to serve as buffers for collision avoidance. A mesh model and a table lookup method were used to quickly identify targets that contain high flight risks. These targets include UAVs, buildings, and restricted airspace. The system would provide appropriate instructions to the operator to avoid the collision. This study utilizes AirSim as the simulated verification platform. Our simulation results show that multiple UAVs can be effectively coordinated, and flight safety issues can be avoided. The maximum computation time can be completed within a 1-second sampling time frame, and this result can be used as a reference for the development of future UAV flight monitoring systems. In the future, we can introduce deep learning and apply decentralized regulatory frameworks to build up the collision avoidance strategy with continuous updating, high performance, and low system fault rate for flight regulation.