Flight Log

1. Jul 24, 2026 · Scobee Flying Field · Flight #0121

   5.4 min · GPS reference available · VIN 4.7 Hz · Mean 1.03 m · Median 1.2 m · Max 3.19 m

   Purpose

   Field validation of VIN absolute position error against GPS ground truth at Scobee Flying Field (AMA community) — quantify locked-mode accuracy and unlock/reacquire behavior over a full sortie.

   Full-resolution file download — available with L1 access.

   Flight report

   

   Full scored table (.xlsx) — available with L1 access.

   Findings

   Locked-mode error: median 1.2 m, mean 1.03 m, max 3.19 m; 99.8% of locked frames ≤ 3 m. Vision lock held ~56% of frames (~861/1530); the dominant unlock path was reacquire_consistency (587 frames), with a smaller share from few_matches. Achieved VIN rate ~4.7 Hz over ~5.4 min. Four GPS-anchor events at ~105 s mean interval.

   Next actions

   Tighten the reacquire_consistency gate and reduce unlock churn so lock duty-cycle rises without inflating locked-mode error.

2. Jul 24, 2026 · Scobee Flying Field · Flight #0120

   5.4 min · GPS reference available · VIN 4.7 Hz · Mean 1.01 m · Median 0.9 m · Max 14.91 m

   Purpose

   Same-day Scobee Flying Field baseline before F0121 — score VIN locked-mode error and lock duty-cycle against GPS ground truth.

   Full-resolution file download — available with L1 access.

   Flight report

   

   Full scored table (.xlsx) — available with L1 access.

   Findings

   Strong lock hold: vision_ok ~93.8% of frames (1429/1524). Locked-mode median 0.9 m, mean 1.01 m, p90 1.49 m; 98.7% ≤ 3 m. Outlier max 14.91 m remains — a short spike, not the typical locked state. VIN ~4.7 Hz over ~5.4 min; three GPS-anchor events. Unlock reasons were mostly few_matches / orb_logged_few_matches (no reacquire_consistency storm as in F0121).

   Next actions

   Add / harden outlier rejection on locked-mode innovation so rare spikes (max ~15 m) are gated without hurting the sub-meter median.

3. Jul 23, 2026 · Ray Brickhouse Flying Field · Flight #0117

   4.2 min · GPS reference available · VIN 4.6 Hz · Mean 0.68 m · Median 0.55 m · Max 3.29 m

   Purpose

   Field validation at Ray Brickhouse Flying Field — score VIN locked-mode absolute error against GPS ground truth on a different AMA community site than Scobee Flying Field.

   Full-resolution file download — available with L1 access.

   Flight report

   

   Full scored table (.xlsx) — available with L1 access.

   Findings

   Best locked accuracy so far in this ledger: median 0.55 m, mean 0.68 m, max 3.29 m; 99.8% of locked frames ≤ 3 m and 76.8% ≤ 1 m. Lock duty-cycle only ~45% (522/1160); dominant unlock path reacquire_consistency (425 frames), then few_matches / innovation_gate. VIN ~4.6 Hz over ~4.2 min; five GPS-anchor events at ~60 s mean interval.

   Next actions

   Same theme as Scobee Flying Field F0121: raise lock duty-cycle by tightening reacquire_consistency without giving up the sub-meter locked median.

4. Jul 21, 2026 · Ray Brickhouse Flying Field · Flight #0109

   4.2 min · GPS reference available · VIN 4.6 Hz · Mean 0.7 m · Median 0.41 m · Max 3.73 m

   Purpose

   Snake-pattern sortie at Ray Brickhouse Flying Field (snake3) — score VIN locked-mode absolute error against GPS ground truth under dense turning / lane changes.

   Full-resolution file download — available with L1 access.

   Flight report

   

   Full scored table (.xlsx) — available with L1 access.

   Findings

   Strong combined result: lock duty-cycle ~96% (1105/1151) with locked-mode median 0.41 m, mean 0.7 m, max 3.73 m; 99% of locked frames ≤ 3 m and 71.4% ≤ 1 m. Unlock noise was low (few_matches 36, reacquire_consistency only 7). VIN ~4.6 Hz over ~4.2 min. GPS-anchor cadence was high (47 fires, ~5.4 s mean interval) — note when comparing to later Ray Brickhouse / Scobee Flying Field runs with sparser anchors.

   Next actions

   Use this snake3 profile as a high-lock baseline; verify the same median holds when GPS-anchor interval is lengthened toward the ~60–105 s regime used on later flights.

VIN engine → · Public Visual Autonomy →