航母甲板上舰载机的混合避障和仿真  

Obstacle Avoidance and Simulation of Carrier-Based Aircraft on the Deck of Aircraft Carrier

作　　者：薛均晓 孔祥燕 董博威 陶浩 管海洋 石磊[1] 徐明亮[2] Xue Junxiao;Kong Xiangyan;Dong Bowei;Tao Hao;Guan Haiyang;Shi Lei;Xu Mingiang(School of Cyber Science and Engineering,Zhengzhou University,Zhengzhou 450002,China;School of Computer and Artificial Intelligence,Zhengzhou University,Zhengzhou 450001,China;China Ship Research and Design Center,Wuhan 430064,China)

机构地区：[1]郑州大学网络空间安全学院,河南郑州450002 [2]郑州大学计算机与人工智能学院,河南郑州450001 [3]中国船舰研究设计中心,湖北武汉430064

出　　处：《系统仿真学报》2023年第3期592-603,共12页Journal of System Simulation

基　　金：国家自然科学基金(62036010,61972362);河南省自然科学基金(202300410378);河南省高等学校青年骨干教师培养计划(22020GGJS014)。

摘　　要：针对航母甲板上舰载机混合避障随机性强、实时性差、规划速度慢等问题,结合最小二乘法与DDPG(deep deterministic policy gradient)算法提出一种PDDPG(predictive depth deterministic policy gradient)算法。该方法利用最小二乘法预测航母甲板上动态障碍物的短期轨迹。DDPG根据动态障碍物的短期轨迹为智能体提供在连续空间里学习和决策行为的能力。基于人工势场设置奖励函数,提高混合避障算法的收敛速度和准确率。使用Unity 3D构建了航母甲板高动态复杂场景,进行舰载机混合避障仿真实验。实验结果表明,PDDPG能较准确地实现航母甲板上舰载机的混合避障,与其他方法相比,在精度上提高了7%~30%。与DQN(deep Q network)相比,路径长度和转弯角度上分别减少了100个单位和400o~450o。A predictive depth deterministic policy gradient(PDDPG)algorithm is proposed by combining the least squares method with deep deterministic policy gradient(DDPG)for the problems of strong randomness,poor real-time performance,and slow planning speed by obstacle avoidance on aircraft carrier deck.The short-term trajectory of dynamic obstacles on the deck is predicted by the least square method.DDPG is used to provide agents with the ability to learn and make decisions in continuous space by the short-term trajectory of dynamic obstacles.The reward function is set based on the artificial potential field to improve the convergence speed and accuracy of the algorithm.A high dynamic complex scene of aircraft carrier deck is constructed using unity 3D to simulate experiments of obstacle avoidance method.The experimental results show that the method can accurately realize the hybrid obstacle avoidance of carrier aircraft on the aircraft carrier deck,and the accuracy is improved by 7%~30%compared with other methods.Compared with deep Q network(DQN),the path length and turning angle are reduced by 100 units and 400º~450ºrespectively.

关 键 词：路径规划 混合避障 最小二乘法 人工势场 DDPG 

分 类 号：TP[自动化与计算机技术]