Robust autonomous landing of UAVs in non-cooperative environments based on comprehensive terrain understanding  

作　　者：Lyujie CHEN Yao XIAO Xiaming YUAN Yiding ZHANG Jihong ZHU 

机构地区：[1]Department of Computer Science and Technology,Tsinghua University,Beijing 100084,China [2]Department of Precision Instrument,Tsinghua University,Beijing 100084,China [3]Department of Automation,TsinghuaUniversity,Beijing100084,China

出　　处：《Science China(Information Sciences)》2022年第11期159-173,共15页中国科学（信息科学）（英文版）

基　　金：supported by National Natural Science Foundation of China(Grant Nos.61903216,62073185)。

摘　　要：Autonomous landing in non-cooperative environments is a key step toward full autonomy of unmanned aerial vehicles(UAVs).Existing studies have addressed this problem by finding a flat area of the ground as the landing place.However,all these methods have poor generalization ability and ignore the semantic feature of the terrain.In this paper,we propose a well-designed binocular-LiDAR sensor system and a robust terrain understanding model to overcome these deficiencies.Our model infers both morphologic and semantic features of the ground by simultaneously performing depth completion and semantic segmentation.Moreover,during the inference phase,it self-evaluates the accuracy of the predicted depth map and dynamically selects the LiDAR accumulation time to ensure accurate depth prediction,which greatly improves the robustness of the UAV in completely unknown environments.Through extensive experiments on our collected low-altitude aerial image dataset and real UAVs,we verified that our model effectively learned two tasks simultaneously and achieved better performance than existing depth estimation-based landing methods.Furthermore,the UAV can robustly select a safe landing site in several complex environments with about 98%accuracy.

关 键 词：UAV autonomous landing terrain understanding depth completion semantic segmentation 

分 类 号：V279[航空宇航科学与技术—飞行器设计] V249