摘要
针对空间机械臂操作过程对卫星非合作结构位姿难以快速准确估计的问题,提出一种基于神经辐射场的卫星非合作关键结构位姿估计方法。本方法首先通过RGBD相机在线获取场景点云,并对点云进行识别与分割得到卫星关键结构,然后利用神经辐射场自动建立关键结构三维模型,最后基于位姿生成网络得到准确的位姿估计。搭建由RGBD深度相机、机械臂和卫星模型组成的试验平台,对不同位姿的卫星关键结构进行位姿估计试验。试验结果表明,所提算法可以在线自动构建非合作目标的三维模型,而无须提前人为准备目标数据,从而实现空间操作中真正意义上的非合作目标位姿估计及跟踪。
Aiming at the problem of difficulty in quickly and accurately estimating the pose of non-cooperative structures of satellites during the operation of space robotic arms,a neural radiation field based method for estimating and tracking the pose of non-cooperative key structures of satellites is proposed.This method first obtains the scene point cloud online through an RGBD camera,segments the point cloud to obtain satellite key structures,and then uses neural radiation fields to automatically establish a three-dimensional model of the key structures.Finally,based on the initial pose generation network and pose evaluation network,accurate pose estimation is obtained.An experimental platform consisting of an RGBD depth camera,a robotic arm,and a satellite model is constructed to conduct pose estimation experiments on key structures of satellites with different poses.The experimental results show that the algorithm proposed can automatically construct a 3D model of non-cooperative targets online without the need for human preparation of target data in advance.At the same time,it can effectively deal with target object occlusion and motion situations,thus achieving true non cooperative target pose estimation and tracking in spatial operations.
作者
陈彦江
王燕波
梁斌焱
林俊钦
CHEN Yanjiang;WANG Yanbo;LIANG Binyan;LIN Junqin(Beijing Institute of Precision Mechatronics and Controls,Beijing,100076;Innovation Center for Control Actuators,Beijing,100076)
关键词
卫星非合作结构
位姿估计
神经辐射场
点云分割
多层感知机
satellite non-cooperative structure
pose estimation
neural radiation field
point cloud segmentation
multilayer perceptions
