火星精确着陆智能融合自主导航方法  

Intelligent Fusion Autonomous Navigation Method for Mars Precise Landing

作　　者：高锡珍 黄翔宇[1,2] 徐超 GAO Xizhen;HUANG Xiangyu;XU Chao(Beijing Institute of Control Engineering,Beijing 100194,China;National Key Laboratory of Space Intelligent Control Technology,Beijing 100194,China)

机构地区：[1]北京控制工程研究所,北京100094 [2]空间智能控制技术全国重点实验室,北京100094

出　　处：《深空探测学报（中英文）》2024年第1期24-30,共7页Journal Of Deep Space Exploration

基　　金：国家自然科学基金(61673057)。

摘　　要：针对未知环境下难以进行绝对光学导航定位的问题,提出了火星精确着陆智能融合自主导航方法。考虑地外环境纹理相似度高、图像间视角尺度变换带来的特征无法检测和识别效率低的困难,构建无监督单应网络估计探测器帧间运动。结合惯性测量信息建立探测器状态递推模型,通过无迹卡尔曼滤波(Unscented Kalman Filter,UKF)利用所建立的测量模型和状态递推模型实现探测器位置、速度和姿态实时估计,仿真结果验证了无需特征检测与匹配,利用视觉惯性测量信息融合导航的有效性。To overcome the difficulty of absolute optical navigation in unknown environments,an intelligent fusion autonomous navigation method for Mars precise landing was proposed.Considering the difficulties of the inability to detect features and the low efficiency of recognition brought by high texture similarity in the extraterrestrial environment and perspective scaling between images,an unsupervised homography network was constructed to estimate the inter frame motion of the lander.Based on the inertial measurement information,a recursive model of the lander state was established.Using the established measurement model and state recursive model,real-time estimation of the lander position,velocity,and attitude was achieved through UKF.The simulation results verify the effectiveness of the proposed method without the need of feature detection and matching.

关 键 词：火星着陆 智能导航 多源融合 无监督学习 深度神经网络 

分 类 号：V488.2[航空宇航科学技术]