Data incest in cooperative localisation with the Common Past-Invariant Ensemble Kalman filter
Data incest in cooperative localisation with the Common Past-Invariant Ensemble Kalman filter
复制标题
使用通用过去不变集成卡尔曼滤波器进行协作定位中的数据乱伦
DOI：
--
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
V. Cahill
中科院分区：
文献类型：
--
作者：
J. Curn;D. Marinescu;N. O'Hara;V. Cahill
在本文中，我们考虑的问题，合作车辆定位，其中一组车辆驾驶在室外环境中，每个估计他们的位置使用全球定位系统（GPS）和里程计。此外，车辆可以通过使用诸如雷达的接近传感器和相互通信来观察其他车辆的位置来改进它们的估计，这对于在没有GPS覆盖的区域中操作的那些车辆特别有帮助。在分布式融合系统中，每个车辆需要考虑这样一个事实，即从其他车辆接收的信息可能部分来自车辆本身，从而导致状态估计和观测误差之间的相关性。这个问题，也被称为数据乱伦，被放大的动态和非结构化的通信拓扑结构的性质，固有的合作本地化方案。我们提供了一种新的解决方案，该问题的基础上共同的过去不变的Envariant卡尔曼滤波器（CPI-EnKF）-一个概括的Envariant卡尔曼滤波器，可以应用在存在共同的过去信息之间共享的状态估计和观察，这是最近提出的本文的作者。正如我们将证明的那样，CPI-EnKF应用更简单，提供更好的估计，可以扩展到任意数量的车辆，并且比其他类似方法计算效率更高。
In this paper we consider the problem of cooperative vehicle localisation, in which a group of vehicles are driving in an outdoor environment, each estimating their position using a global positioning system (GPS) and odometry. Additionally, the vehicles can improve their estimates by observing positions of other vehicles using a proximity sensor, such as a radar, and a mutual communication, which is especially helpful to those vehicles operating in areas with no GPS coverage. In a distributed fusion system, each vehicle needs to account for the fact that information received from other vehicles might originate in part from the vehicle itself, resulting in a correlation between the state estimate and observation errors. This problem, also known as data incest, is amplified by the dynamic and unstructured nature of the communication topology, inherent to a cooperative localisation scenario. We provide a novel solution to the problem based on the Common Past-Invariant Ensemble Kalman filter (CPI-EnKF) - a generalisation of the Ensemble Kalman filter that can be applied in the presence of common past information shared between the state estimate and the observation, which has been recently proposed by this paper's authors. As we will demonstrate, the CPI-EnKF is simpler to apply, provides better estimates, can be scaled to an arbitrary number of vehicles and is computationally more efficient than other similar methods.