Data-driven optimal shared control of unmanned aerial vehicles

作     者：Tan, Junkai Xue, Shuangsi Guo, Zihang Li, Huan Cao, Hui Chen, Badong 

作者机构：Xi An Jiao Tong Univ Sch Elect Engn Xian 710049 Peoples R China Xi An Jiao Tong Univ Sch Elect Engn State Key Lab Elect Insulat & Power Equipment Xian 710049 Peoples R China Xi An Jiao Tong Univ Natl Engn Res Ctr Visual Informat & Applicat Natl Key Lab Human Machine Hybrid Augmented Intell Xian 710049 Peoples R China Xi An Jiao Tong Univ Inst Artificial Intelligence & Robot Xian 710049 Peoples R China 

出 版 物：《NEUROCOMPUTING》 (Neurocomputing)

年 卷 期：2025年第622卷

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Founda-tion of China [62436005, 62311540022] China Postdoctoral Science Foundation China [2024M762602] 

主　　题：Shared control Reinforcement learning Optimal control Koopman operator Unmanned aerial vehicle Approximate dynamic programming 

摘      要：Cooperation between humans and autonomy is a critical topic of unmanned aerial vehicle (UAV) control. How to co-pilot the UAV with human operator to achieve optimal performance presents a significant challenge. In this paper, we propose a novel data-driven optimal shared control method for UAV using the Koopman operators to predict the nonlinear dynamics of the UAVs. An original shared control mechanism is established to allocate the relationship between optimal and human control inputs. The model of the system is learned from human maneuver data via the Koopman operator approach, and the optimal controller is approximated online using reinforcement learning techniques. The Lyapunov theory analyzes the stability of the proposed method. Compared with offline RL methods, the proposed method can learn the optimal controller online without a precise UAV dynamics model from human maneuver data. The effectiveness of the proposed method is demonstrated by numerical and Human-in-the-loop (HiTL) simulation.