Fault-tolerant LiDAR-Visual-GNSS Localization Using a Split Covariance Intersection Filter

To overcome the limitations of conventional sensor-fusion methods that are susceptible to sensor and algorithmic failures, this study proposes a fault-tolerant localization method that integrates LiDAR, camera, GNSS, and IMU. The proposed method employs a hierarchical fusion architecture centered on a Health-Gated Split Covariance Intersection Filter (HG-SCIF), which dynamically reconfigures the fusion set based on the health status of each localization module while accounting for unknown cross-correlations among lower-level estimates. Coordinate transformation and pose smoothing are also incorporated to facilitate seamless transitions among available sensor combinations. This design enables continuous localization using the sensors that remain functional when partial system failures occur. Simulation experiments conducted using the MORAI simulator demonstrate that the proposed system detects and excludes faulty localization modules and dynamically reconfigures the fusion set, while maintaining stable and continuous localization estimates under both sequential and simultaneous fault-injection scenarios.