Control-oriented UAV highly feasible trajectory planning: A deep learning method

被引:41
作者
Liu, Yiheng [1 ,2 ,3 ]
Wang, Honglun [1 ,3 ]
Fan, Jiaxuan [4 ]
Wu, Jianfa [1 ,2 ,3 ]
Wu, Tiancai [1 ,2 ,3 ]
机构
[1] Beihang Univ, Sch Automat Sci & Elect Engn, Beijing 100191, Peoples R China
[2] Beihang Univ, Shenyuan Honors Coll, Beijing 100191, Peoples R China
[3] Beihang Univ, Sci & Technol Aircraft Control Lab, Beijing 100191, Peoples R China
[4] China Acad Launch Vehicle Technol, Res & Dev Ctr, Beijing 100071, Peoples R China
基金
中国国家自然科学基金;
关键词
Highly feasible trajectory planning; Trajectory-mapping network; Deep learning; Unmanned aerial vehicle; UNMANNED AERIAL VEHICLES; PATH; NETWORKS; TRACKING; TARGET;
D O I
10.1016/j.ast.2020.106435
中图分类号
V [航空、航天];
学科分类号
08 ; 0825 ;
摘要
The highly feasible trajectory planning of unmanned aerial vehicle (UAV) is very important in some tasks but has not yet attracted sufficient study attention. Most current studies use simplified UAV model with some state constraints to plan the trajectory, but the feasibility is reduced, because the simplified model is very different from the actual UAV system, so that the tracking characteristics of UAV cannot be fully considered. In this paper, a novel control-oriented UAV highly feasible trajectory planning method is proposed. First, a UAV closed-loop model prediction method, which is the combination of a low-level controller and a UAV 6 DOF nonlinear model, is adopted in the trajectory planning phase to predict the flight trajectory. This complicated model is very similar to the actual UAV system because it comprehensively considers the controller performance and the detailed UAV model, but it also has poor efficiency. Therefore, a trajectory-mapping network (TMN) is proposed using a deep learning approach to improve the planning efficiency. Furthermore, a novel time-series convolutional neural network (TSCNN) is proposed for the TMN to further improve its computation speed and prediction accuracy. Finally, the flight trajectory predicted by the TMN is used to evaluate the planning cost. In this way, the planned trajectory will be highly feasible. The effectiveness of the proposed method is demonstrated by simulations. (C) 2020 Elsevier Masson SAS. All rights reserved.
引用
收藏
页数:11
相关论文
共 28 条
[1]   Suboptimal artificial potential function sliding mode control for spacecraft rendezvous with obstacle avoidance [J].
Cao, Lu ;
Qiao, Dong ;
Xu, Jingwen .
ACTA ASTRONAUTICA, 2018, 143 :133-146
[2]   UAV Trajectory Optimization for Data Offloading at the Edge of Multiple Cells [J].
Cheng, Fen ;
Zhang, Shun ;
Li, Zan ;
Chen, Yunfei ;
Zhao, Nan ;
Yu, F. Richard ;
Leung, Victor C. M. .
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, 2018, 67 (07) :6732-6736
[3]  
Hochreiter S., 1997, Neural Computation, V9, P1735
[4]   MULTILAYER FEEDFORWARD NETWORKS ARE UNIVERSAL APPROXIMATORS [J].
HORNIK, K ;
STINCHCOMBE, M ;
WHITE, H .
NEURAL NETWORKS, 1989, 2 (05) :359-366
[5]  
Huang Y., 2018, IEEE T AEROSP ELECT
[6]   A method of 3D path planning for solar-powered UAV with fixed target and solar tracking [J].
Huang, Yu ;
Chen, Jianguo ;
Wang, Honglun ;
Su, Guofeng .
AEROSPACE SCIENCE AND TECHNOLOGY, 2019, 92 :831-838
[7]  
Ioffe Sergey, 2015, PMLR, P448
[8]   Optimal path planning for unmanned combat aerial vehicles to defeat radar tracking [J].
Kabamba, PT ;
Meerkov, SM ;
Zeitz, FH .
JOURNAL OF GUIDANCE CONTROL AND DYNAMICS, 2006, 29 (02) :279-288
[9]   Optimal maneuver-based motion planning over terrain and threats using a dynamic hybrid PSO algorithm [J].
Karimi, J. ;
Pourtakdoust, Seid H. .
AEROSPACE SCIENCE AND TECHNOLOGY, 2013, 26 (01) :60-71
[10]  
Kim E., 2007, CONTROL SIMULATION R