基于视觉导航与强化学习的可重复使用火箭垂直降落控制策略研究  

Research on vertical landing control strategy of reusable rocket based on visual navigation and reinforcement learning

作　　者：尹超 Yin Chao(Shanghai Aviation Electric Co.,Ltd.,Shanghai 200030,China)

机构地区：[1]上海航空电器有限公司,上海200030

出　　处：《无线互联科技》2023年第7期141-143,共3页Wireless Internet Technology

摘　　要：随着商业航天在我国的快速发展,星链计划体现的经济和国防作用越来越明显,可重复使用火箭垂直降落方式是商业航天中决定商业成功的关键因素之一。考虑到商业火箭在重复使用过程中具备抗干扰和低成本的特点,文章提出了采用纯视觉与强化学习融合的方法。文章通过多角度视觉传感融合实现火箭定位并强化学习所规划的航路,确保火箭在可控飞行能力范围以内精准垂直降落在指定着陆区域。针对火箭垂直降落路径规划与控制精度的难点,文章提出了基于火箭动力学特性和飞行数据的动力学参数辨识方法,为提高火箭垂直降落运动轨迹预估精度提供有效手段。With the rapid development of commercial aerospace in China,the economic and national defense role of the Star Chain program is becoming increasingly evident.The vertical landing method of reusable rockets is one of the key factors determining commercial success in commercial aerospace.Considering that commercial rockets have the characteristics of anti-interference and low cost in the process of repeated use,this paper proposes a fusion method of pure vision and reinforcement learning.Rocket positioning is achieved through multi-angle visual sensor fusion,and learning the planned route is enhanced to ensure that the rocket accurately and vertically lands in the designated landing area within the controllable flight capability range.Aiming at the difficulties in the planning and control accuracy of rocket vertical landing path,a dynamic parameter identification method based on rocket dynamic characteristics and flight data is proposed,which provides an effective means to improve the prediction accuracy of rocket vertical landing trajectory.

关 键 词：视觉导航 强化学习 系统辨识 控制策略 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]