SINS/GNSS车载组合导航自适应抗差滤波算法

Adaptive Robust Filtering Algorithm for Vehicle-Borne SINS/GNSS Integrated Navigation

摘要: 针对城市车载场景中GNSS信号遮挡、多路径粗差与时变非平稳噪声，标准ESKF存在线性近似误差累积问题，传统UKF采样冗余、无自适应噪声调节与粗差抑制能力，本文提出比例采样自适应抗差无迹卡尔曼滤波(PS-ARUKF)。基于SINS十五维误差状态模型与GNSS观测模型，算法设计比例修正最小Sigma点采样结构，削减矩阵运算量，弥补ESKF非线性拟合不足，采用指数加权改进Sage-Husa机制动态更新噪声协方差，适配车载突变噪声工况，引入Huber抗差因子降低异常观测权重，提升滤波鲁棒性。开展闭合路径实车重复性试验，对比标准ESKF、传统UKF与PS-ARUKF性能。实测结果表明，相较传统UKF，PS-ARUKF水平位置RMSE精度提升48.60%，终点位置误差精度提升40.09%，航向角误差精度提升31.90%，单步滤波平均耗时0.20 ms，实时性能接近工程主流ESKF，可在兼顾运算实时性的同时显著提升导航精度，满足智能驾驶复杂路况下车载高精度组合导航应用需求。

Abstract: Aiming at GNSS signal occlusion, multipath gross errors and time-varying non-stationary noise in urban vehicle-borne scenarios, the standard ESKF suffers from accumulated linear approximation errors, while the traditional UKF has redundant sampling points without adaptive noise adjustment and gross error suppression capabilities. This paper proposes a Proportional Sampling Adaptive Robust Unscented Kalman Filter (PS-ARUKF). Based on the 15-dimensional SINS error state model and GNSS observation model, the algorithm designs a proportionally modified minimum Sigma-point sampling structure to reduce matrix computation and compensate the insufficient nonlinear fitting capacity of ESKF. An exponentially weighted improved Sage-Husa mechanism is adopted to dynamically update noise covariance for adapting to sudden vehicle-borne noise conditions, and the Huber robust factor is introduced to reduce the weight of abnormal observations and improve filtering robustness. Repeated real vehicle experiments on closed paths are carried out to compare the performance of standard ESKF, traditional UKF and PS-ARUKF. The test results show that compared with the traditional UKF, the horizontal position RMSE accuracy of PS-ARUKF is improved by 48.60%, the terminal position error accuracy by 40.09%, and the heading angle error accuracy by 31.90%. Its average single-step filtering time is 0.20 ms, with real-time performance close to the mainstream engineering ESKF. The proposed algorithm can significantly improve navigation accuracy while guaranteeing computational real-time performance, and meet the application requirements of high-precision vehicle-borne integrated navigation for intelligent driving under complex road conditions.