NavSol

Navigation and state-estimation work on the INSANE multi-sensor UAV dataset: dataset characterization, ground-truth verification, a corrected ground-truth smoother, and a navigation-state estimator.

NavSol is built on the INSANE cross-domain multi-sensor UAV dataset [1]. The work proceeds in four stages, each a section below.

INSANE Dataset Guide

INSANE Dataset Guide characterizes the dataset for navigation-state estimation: platform and hardware invariants, frames and clocks, the per-stream sensor reference, calibration, source of truth, data quality, and per-run detail. Every quantitative claim is verified against the distributed files.

Ground-Truth Verification

Ground-Truth Verification reproduces the maintainers’ shipped ground truth from the raw CSVs and characterizes where its attitude is gravity-inconsistent.

Corrected Ground-Truth Smoother

Corrected Ground-Truth Smoother is the corrected ground-truth-like product across all 26 flights, together with its computer-vision feasibility, visual-inertial bundle-adjustment, and camera–LRF calibration dependencies.

Navigation-State Estimator

Navigation-State Estimator is the estimator, scored against both the smoother product and the maintainer ground truth.

References

[1]

C. Brommer et al., “INSANE: Cross-domain UAV data sets with increased number of sensors for developing advanced and novel estimators,” Oct. 2022, doi: 10.48550/arXiv.2210.09114.