Robust Kalman Filter Enhanced by Projection Statistic Detector for Multisensor Navigation in Urban Canyon Environment

• Published in: IEEE sensors journal 2023-05, Vol.23 (9), p.9832-9847
• Main Authors: Jiang, Yingying, Pan, Shuguo, Meng, Qian, Gao, Wang, Ma, Chun, Yu, Baoguo, Jia, Fengshuo
• Format: Article
• Language: English
• Subjects: Algorithms; Autonomous cars; Availability; Dynamic tests; Extended Kalman filter (EKF); Global navigation satellite system; huber’s M-estimation; Iterative methods; Kalman filters; loosely couple; multisensor integration; Navigation systems; Particle measurements; Pollution measurement; projection statistics (PSs); Redundancy; Regression models; Robustness; Robustness (mathematics); Safety critical; Sensors; State estimation; Statistical analysis; Street canyons; urban canyon; Urban environments
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2022.3230708

Multisensor navigation via redundancy and complementary has been widely applied to safety-critical services, such as self-driving vehicles. To improve the robustness and reliability of the multisensor navigation system in urban canyon environments, a robust Kalman filter (RKF) method enhanced by the projection statistic (PS) detector is proposed in this work. Based on a statistical consistency check, the availability of measurements from observation sensors is preevaluated by the PS criterion. The new iterative Huber's M-estimation with the exclusion function is implemented on the linear regression model for robust state estimation. The effectiveness of the proposed algorithm was verified by a dynamic test in the representative urban canyon environment. Our approach shows significant superiority and robustness among the comparative experiments, where the 3-D root-mean-square (rms) error is limited to 3.33 m. The biased measurements can be effectively identified and removed by the PS detector with a given significance level. The new iterative Huber's M-estimation assisted by the prior availability knowledge enables the integration solution more robust and reliable. The presented robust method is pretty suitable for multisensor navigation in urban canyon environments.