GNSS/惯性/视觉组合导航时间序贯完好性监测方法研究

保障多传感器组合导航的完好性是实现自主系统安全运行的前提。对于基于滤波的信息融合方法，其完好性表现高度取决于测量误差的大小和时间自相关性，然而，这些误差特性难以从数学上精确描述，这给完好性设计带来了挑战。本项目针对全球导航卫星系统（GNSS）、惯性导航系统（INS）与视觉里程计（VO）组合的导航系统中存在的测量故障与误差随机模型不确定性，拟基于奇偶矢量对不同故障模式的空间投影关系提出故障排除算法，推导时间相关常数不定条件下的保护水平包络公式，以严谨地量化安全风险，探索时间序贯完好性监测理论。针对VO测量易受场景影响且故障率高的问题，提出多级检测算法，建立路标点误差紧包络模型。针对INS零偏以及GNSS多个误差源的自相关时间不确定问题，基于时间常数上下边界提出保守的标准差估算方法。本项目拟通过系统仿真和无人机飞行实验验证理论正确性，预期成果将为提高有视觉深度参与的组合导航安全性提供理论支持。

Ensuring high integrity of multi-sensor integrated navigation system is the precondition to achieve safe operation of autonomous systems. For filter-based information fusion, the integrity performance is highly dependent on the magnitude and time correlation profiles of the measurement error. However, precise mathematical nature of the error characteristics is rarely known, which causes significant challenges to integrity design. This project aims at rigorously quantifying the safety risk against measurement faults and stochastic error modeling uncertainty in an integrated navigation system using Global Navigation Satellite Systems (GNSS), Inertial Navigation System (INS) and Visual Odometry (VO). The integrity monitoring is achieved by (a) proposing a fault exclusion algorithm based on the relationships of the parity vector projections on different fault modes, (b) deriving conservative protection level equations with unknown correlation time constants, and (c) establishing an analytical method to evaluate filter-based time sequential integrity risk. To mitigate the impact of operational scenario changes on VO measurements, and to limit the failure rate, a multi-layer detection scheme is proposed, which is followed by an overbounding error model establishment. Moreover, a conservative standard deviation evaluation method is derived based on the upper and lower time constant bound, which resolves the autocorrelation uncertainty issue of INS bias and multiple GNSS errors. The proposed scheme is realized and validated through system simulations and Unmanned Aerial Vehicle (UAV) flight test. The anticipated outcomes of this project will provide theoretical support for improving VO-participated multi-sensor integrated navigation safety.