Ground-Truth Verification
Every claim here was checked directly against the distributed INSANE dataset (per-domain verification across IMU, magnetometer, baro, PX4 GPS, RTK, LRF, UWB, RealSense odometry, motor, mocap, ground truth, Mars state exports, camera/vehicle/magnetometer calibration, timing, images, and raw bundles, with an adversarial audit pass).
These foundations studies underpin corrections in Motion Capture, Calibration, Source of Truth, Contradictions, What Is Not in the Download, and Per-Run Reference:
- Reproducing the shipped ground truth (code):
- regenerates the shipped GT bit-for-bit (raw
px4_mag; 18/19 Mars missions to ~1e-15 m / ~1e-6°) - establishes the Klagenfurt-site GT non-reproducibility
- the raw-mag confirmation
- regenerates the shipped GT bit-for-bit (raw
- Magnetometer intrinsic calibration (code) — reproduces the mag intrinsic to ~1e-13 and the LSM site-swap defect.
- Ground-truth attitude: gravity inconsistency (code):
- the static-epoch gravity-consistency check on the shipped GT attitude (0.9–55° median, ≤~77°)
- computed on the shipped
px4_imu.csv+ground_truth_80hz.csvdirectly for all 19 Mars missions and the four KLU-site runs - by two independent code paths
- Is it the magnetometer intrinsic? (code):
- regenerates the Wahba GT with the published
px4_magintrinsic applied (vs raw) and shows it does not restore gravity-consistency at dual-RTK-fixed epochs (mean tilt 8.0°→8.5°, 9/19 improve, near-correct missions degrade) - proving the inconsistency is structural
- also documents the RTK-fix-quality decomposition (float collapses the baseline → tilt)
- regenerates the Wahba GT with the published
- Transition-handoff reconstruction — the marker-based transition-handoff reconstruction (board geometry, ArUco regeneration, cross-checks) and its adversarial audit.