【摘要】 为解决移动机器人组合导航系统中由于存在时变、非高斯噪声而导致的估计精度下降问题,提出一种将秩卡尔曼滤波器(Rank Kalman Filter, RKF)与Huber统计线性回归近似方法相结合的Huber秩卡尔曼滤波算法(Huber-RKF)。RKF与高斯确定点采样型滤波算法结构相似,但它不需要满足高斯分布假设条件,完全利用秩统计量相关原理计算采样点及其权值,适用于非线性、非高斯系统;Huber方法将l1/l2混合范数作为代价函数,通过迭代求得最优估计值,具有较好的鲁棒性;把二者相结合,将Huber最优估计作为RKF算法结构中的量测更新,得到的Huber-RKF算法具有良好的鲁棒性和滤波估计精度。仿真实验中将Huber-RKF与EKF、RKF以及交互式多模型秩卡尔曼滤波器(IMM-RKF)进行比较,其纬度、经度估计误差分别减小了69.5%、75.6%,44%、44.1%,27%、14%;算法实时性方面,Huber-RKF算法中程序循环体单次执行的时间为20.8 ms,比IMM-RKF执行速度快33%。更多还原

【Abstract】 In tough environment, the noise of the integrated navigation system for mobile robot tends to be time-varying and non-Gaussian. To tackle this problem, a novel filter named Huber-based rank Kalman filter(Huber-RKF) is proposed, which combines a rank Kalman filter(RKF) with Huber’s generalized maximum likelihood approach into the estimation. The RKF’s structure is similar with that of the Gaussian fixed-point sampling filtering algorithm, but it does not need to meet the assumption of Gaussian distribution, and fully uses the rank statistic correlation principle to calculate the sampling points and their weights, which is applicable to non-linear and non-Gaussian systems. Huber method takes l1/l2 mixed norm as the cost function and obtains the optimal estimate through iteration, which has good robustness.Replacing the measurement update step of RKF algorithm structure with Huber’s method will significantly increase the robustness and accuracy of the algorithm. Simulation results show that th e Huber-RKF algorithm could respectively improve the precisions of latitude and longitude estimation by 69.5% and75.6% compared to EKF, by 44% and 44.1% compared to RKF, and by 27% and 14% compared to IMM-RKF. The Huber-RKF can reduce the single execution time of program loop to 20.8 ms, which is 33%faster than that of the IMM-RKF.更多还原