Pilot-residual-driven adaptive robust fusion of inertial and Doppler velocity measurements for underwater navigation
Doppler velocity log (DVL) observations constrain inertial drift in underwater navigation but may exhibit changing dispersion, impulsive errors and temporary loss. We examine a two-stage batch factor-graph estimator that uses posterior DVL residuals from a Gaussian pilot to construct separate scale and standardized-tail statistics. The former modulates DVL factor scale; the latter selects a generalized robust weight. The graph uses calibrated IMU– DVL geometry and preintegrated inertial measurements, and omits missing DVL factors. On two held-out trajectories under 96 injected-fault realizations, mean trajectory error was 1.128 m versus 1.196 m for Gaussian and 1.186 m for pilot-weighted Huber. The paired difference from a pilot-weighted fixed Geman–McClure-type baseline was not significant. Without retuning, transfer to the independent TANK dataset yielded a mean error of 2.837 m across 96 further perturbations, with a 6.09% paired gain over the Geman-type baseline under impulsive errors. Ablations identify tail adaptation, rather than broad-noise scaling, as the principal supported component. The evidence concerns controlled perturbations of four physical sequences, not independent missions, and does not establish real-time operation or calibrated uncertainty.
Authors
- Seyyede Marzieh Mousavi
- Sobhan Toulabi
Institutions
- Amirkabir University of Technology (IR)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-19
- DOI
- https://doi.org/10.21203/rs.3.rs-11540344/v1
- Primary Topic
- Inertial Sensor and Navigation
- Type
- preprint