【作者】 王小刚；

【导师】 崔乃刚；

【作者基本信息】 哈尔滨工业大学 ， 飞行器设计， 2010， 博士

【摘要】 “协同编队飞行”是一种全新的无人机应用模式,具有广阔的前景和巨大的技术优势,代表了无人机技术的未来发展方向,因而备受关注。本文以无人机协同编队飞行为背景,对编队无人机之间的相对导航问题展开研究,通过对多种相对导航方案的对比,最终确定了惯导系统和VisNav视觉导航系统所组成的INS/VisNav相对导航系统方案,并设计了多种相对导航滤波器,主要的内容包括:针对“长机——僚机”编队模式,推导了无人机之间的相对运动模型,给出了陀螺、加速度计和VisNav视觉导航系统的测量模型,对VisNav视觉导航系统在单矢量、双矢量、三矢量以及多矢量观测情况下可观测性进行了分析,为确定VisNav视觉导航系统的工作条件提供了理论依据。采用最小二乘法处理VisNav视觉导航系统的输出数据,估计出编队无人机之间的相对姿态和相对位置。数学仿真分析显示,最小二乘法的估计精度受无人机之间的相对位置关系所影响。设计了基于扩展卡尔曼滤波算法的INS/VisNav相对导航滤波器。用长机和僚机的惯性导航系统测量数据代替相对运动方程中的角速度和加速度项,获得长机和僚机之间的相对惯导方程,以相对惯导方程为基础,推导了扩展卡尔曼滤波中的雅可比矩阵和测量敏感矩阵。考虑到长机与僚机处于编队飞行模式中,忽略重力项影响,从而简化了扩展卡尔曼滤波算法,提高了运算速度。为了改善收敛性能,并提高估计精度,设计了基于Sigma-Point卡尔曼滤波算法的INS/VisNav相对导航滤波器。介绍了Unscented卡尔曼滤波算法和中心差分卡尔曼滤波算法,通过对比分析上述两种算法发现,二者的共同点是利用Sigma-Point近似非线性函数。在设计Sigma-Point卡尔曼滤波器时,考虑到姿态四元数的归一化限制,采用罗德里格斯参数代表姿态误差状态,保证了误差协方差矩阵的非奇异性,仿真结果显示,算法具有明显的快速收敛性能。考虑到当系统噪声不符合高斯分布时,卡尔曼滤波难以取得高精度估计,甚至发散,设计了基于Huber-Based滤波的INS/VisNav相对导航滤波器,Huber-Based滤波器是一种鲁棒滤波器,建立在l1、l2混合范数最小的基础上,当系统噪声为受污染的高斯分布时,该滤波方法具有一定的鲁棒性。将Huber-Based滤波与Sigma-Point滤波相结合,提出了一种鲁棒Sigma-Point滤波方法,并设计了基于鲁棒Sigma-Point滤波的相对导航滤波器,最后进行了数学仿真研究。精度和可靠性是衡量相对导航系统性能的两个重要准则,本文提出了一种INS/VisNav/GPS多传感器信息融合相对导航系统,在分布式融合结构下,采用信息滤波算法融合INS、VisNav和GPS载波相位差分系统输出,与INS/VisNav相对导航系统相比,该系统实现了精度和可靠性的双重提高。更多还原

【Abstract】 Cooperative formation flight is a kind of new application method of Unmanned Aerial Vehicle (UAV) which has a wide prospect and huge technical advantage and represents the future developing direction. Therefore, many researchers pay attention to cooperative formation flight. This paper whose background is cooperative formation flight does research on relative navigation between UAVs in formation. By contrast with multiple relative navigation schemes, INS/VisNav is chosen as the project whose core equipment is VisNav. In the same time, various relative navigation filters are designed. The major contents are as follows:On the basis of introducing the attitude description and attitude kinematics of UAV, according to the“leader-follower”formation mode, the relative kinematics model is derived, the measurement models of gyro, accelerometer and VisNav are provided, and the observability of VisNav is analyzed on the condition of single vector, double vectors, three vectors and multiple vectors measurements. Therefore, the working premise of VisNav could be determined. The least square algorithm is applied to process the data of VisNav to estimate the relative attitude and position. The results of simulation show that the accuracy of least square algorithm would be influenced by the relative position of UAVs in formation.The INS/VisNav relative navigation filter is designed which is based on extended Kalman filtering (EKF). The INS measurement dates are used to substitute the angle velocity and acceleration in relative motion equations. As a result, the relative INS equations could be acquired. In the same time, the Jacobian matrix and measurement matrix of EKF could be derived. In the premise of formation flight of leader and follower, the effect of gravity could be ignored. Therefore, the algorithm is simplified and the calculation velocity is promoted.To improve the convergence rate and estimated accuracy, the relative navigation filter is designed which is based on Sigma-Point Kalman filtering. Firstly, the Unscented Kalman filtering and Central Differential Kalman filtering are introduced. By contrast the two algorithms, the common points could be found which are that the Sigma-Point is used to approximate the nonlinear equations. As a result, the Sigma-Point Kalman filtering is summarized and concluded. For the quaternion normalization, an unconstrained three-component vector is used to presented an attitude error quaternion which could confirm the non-singular of error covariance matrix. The simulation results show that the algorithm has evident quick convergence rate.Considering that when the system noise follows non-Gaussian probability distribution, the Kalman filtering couldn’t estimate accurately, filter divergence would be possible to happen on extreme condition. The relative navigation filter is designed which is based on Huber-Based filtering that is a combined minimum l1 and l2 norm estimator which could exhibit robustness for the perturbed Gaussian probability distribution. The Huber-Based filtering and Sigma-Point Kalman filtering are combined. As a result, a new algorithm which is called robust Sigma-Point Kalman filtering is presented. The relative navigation filter is designed which is based on robust Sigma-Point Kalman filtering. Finally, the simulation was done.The accuracy and reliability are two important standards which are used to evaluate the relative navigation system. This paper presents the INS/VisNav/GPS relative navigation system. Using the hierarchically decentralized structure, the information filtering is applied to fuse the dates from INS, VisNav and GPS system. Contrast with INS/VisNav, this system could promote the accuracy and reliability.更多还原