Derkachi e2e Tier-2 divergence had three stacked root causes; this
commit ships fixes for all three plus the IMU prerequisite they
depend on, plus a baseline cheirality gate for cv2.recoverPose.
AZ-918 MAVLink IMU adapters now convert raw mG/mrad-s + FRD body to
SI m/s^2 + rad/s + FLU body via helpers.imu_units. Without
this the ESKF receives values ~1000x too small with wrong-
sign Y/Z and cannot function at all.
AZ-919 Composition root wires EskfNominalAltitudeProvider into the
KLT/RANSAC strategy via the AZ-331 factory introspect path;
OKVIS2 and VINS-Mono are unaffected.
AZ-920 KLT/RANSAC recovers metric translation via Ground Sampling
Distance when AGL is available; otherwise falls through with
scale_quality=direction_only/unknown (no fake scale invented).
AZ-921 VioOutput.scale_quality signal; ESKF add_vio adapts R_meas
position block based on the flag (1e6 inflation when scale is
direction_only/unknown to keep the filter consistent).
AZ-922 KLT/RANSAC cheirality gate rejects single-frame rotations
beyond a config threshold (default 30 deg), catching
cv2.recoverPose twisted-pair flips that cause immediate ESKF
divergence on low-parallax aerial scenes.
Verification:
- Tier-1 (macOS) unit suite: 2346 passed, 0 failed.
- Tier-2 (Jetson) Derkachi e2e: divergence moves from frame 5
(mahalanobis^2 3757) to frame 233 (mahalanobis^2 212). Remaining
drift is open-loop attitude accumulation, not cheirality.
Follow-up tickets filed:
- AZ-923 closed as misdiagnosed: EskfNominalAltitudeProvider was
already correct (nominal_pos.z IS the AGL when takeoff origin sits
at ground level); the early-frame AGL near zero reflects the drone
being stationary on the ground, not a provider bug.
- AZ-942 filed: cross-check VIO rotation against IMU preintegrator
(consistency gate) - more physically grounded than the coarse
AZ-922 threshold and likely required to absorb the frame-233 drift.
Co-authored-by: Cursor <cursoragent@cursor.com>
Replaces the tlog two-clock replay surface with a single-clock path
driven by the Derkachi-schema CSV. --imu is the new required CLI arg;
--tlog stays as a deprecated alias (warned + ignored when --imu set)
until AZ-895 deletes it.
* csv_ground_truth.py parses the 15-column schema, fails fast at
startup on every documented schema fault (AC-5).
* CsvReplayFcAdapter slots into ReplayInputBundle.fc_adapter alongside
the tlog sibling; mirrors Invariant-5 outbound wiring; inbound bus is
intentionally a no-op since the loop reads CSV directly.
* _run_replay_loop branches on imu_csv_path, stamps
VioOutput.emitted_at_ns from the CSV-derived frame_end_ns (AC-4),
closing the AZ-848 two-clock surface for the new path.
* AZ-896 ships the operator-facing format spec at
_docs/02_document/contracts/replay/csv_replay_format.md plus a
20-row example CSV (AC-3 regression-locked).
Tests: 11 + 12 new unit tests, plus updates to AZ-401 import-boundary
and AZ-402 CLI suites. Full unit suite 2,327 passed / 86 skipped.
Co-authored-by: Cursor <cursoragent@cursor.com>
Cycle-3 refactor run 02-az507 (RouteSpec relocation + module-layout
refresh + AZ-270 lint widening). Single batch of 3 tasks; epic AZ-844.
AZ-845 — Relocate RouteSpec DTO to _types/route.py (rule-9 fix):
* New canonical home: src/gps_denied_onboard/_types/route.py
(frozen+slots dataclass; full docstring carried over verbatim).
* c11_tile_manager/route_client.py imports from _types.route.
* replay_input/tlog_route.py and replay_input/__init__.py keep
re-exports for backward-compat (RouteSpec in __all__).
* 5 test files updated to import from _types.route for symmetry.
* Identity-preserving re-export verified by new test
test_az845_routespec_canonical_home_and_reexport_identity.
AZ-846 — Refresh module-layout.md cycle-3 entries:
* c11_tile_manager Internal list rewritten with all 8 internals
(alphabetised) — corrects a stale entry that referenced files
(satellite_provider_*.py) that no longer exist.
* shared/replay_input file list adds errors.py (cycle-2 carry),
tlog_ground_truth.py (cycle-2 carry), tlog_route.py (cycle-3 NEW).
* shared/_types section registers route.py with provenance line.
* Out-of-scope cycle-2 carry-overs (replay_api/, cli/render_map.py,
helpers/gps_compare.py, etc.) intentionally untouched.
AZ-847 — Widen test_az270 lint to enforce full rule-9 allow-list:
* test_ac6_only_compose_root_imports_concrete_strategies now walks
every components/<X>/*.py ImportFrom/Import and rejects anything
not in the rule-9 allow-list (own subpackage + _types + helpers
+ config/logging/fdr_client/clock + frame_source interface-only).
* Strict superset of the original AC-6 narrow check.
* Reports zero violations on the codebase post-AZ-845.
* Two principled carve-outs documented in the test docstring:
- components/<X>/bench/** path skip (measurement code legitimately
constructs production strategies via runtime_root factories).
- register_* lazy self-registration imports from
runtime_root.<X>_factory (central-registry plugin pattern).
* Both carve-outs surfaced to user via Choose A/B/C/D Risk-1
protocol; user skipped both — agent proceeded with documented
defaults. Doc-only follow-up tracked in
_docs/_process_leftovers/2026-05-24_az847_rule9_wording_followup.md
for rule-9 wording update in module-layout.md.
Test results: 2287 passed, 90 skipped (environmental — Docker / CUDA
/ TensorRT / Jetson hardware / fixtures), 0 failed. Focused subset
(replay_input/ + c11_tile_manager/ + test_az270_compose_root.py)
also clean: 169 passed, 1 skipped.
Tracker: AZ-845/846/847 transitioned In Progress -> In Testing.
Co-authored-by: Cursor <cursoragent@cursor.com>
First building block of Epic AZ-835. Pure function that consumes
an ArduPilot binary tlog and returns a RouteSpec (waypoints +
per-waypoint coverage radius + provenance) suitable for posting
to satellite-provider's POST /api/satellite/route endpoint.
Pipeline:
- Load GPS fixes via existing load_tlog_ground_truth (AZ-697).
- Trim leading + trailing rows below takeoff thresholds
(speed >= 2 m/s AND AGL >= 5 m by default; configurable).
- Coarsen to <= max_waypoints via iterative Douglas-Peucker on
the local-ENU projection (WgsConverter.latlonalt_to_local_enu,
AZ-279). DP tolerance is caller-supplied or binary-searched
(<= 32 iterations, <= 1 m convergence).
Public surface (re-exported from replay_input/__init__.py):
- RouteSpec (frozen, slots, with provenance fields).
- RouteExtractionError (subclass of ReplayInputAdapterError).
- extract_route_from_tlog().
Tests: 14 unit tests cover AC-1..AC-10 plus edge cases (custom
DP tolerance, invalid inputs, error hierarchy, too-short segment).
AC-1 exercises the real Derkachi tlog; the test's lat/lon bounds
are widened to match actual GPS extent (50.0800..50.0840 /
36.1070..36.1145) — the AZ-836 spec's tighter IMU-derived bounds
(50.0808..50.0832 / 36.1070..36.1134) cover only the IMU-active
window, not GPS-active takeoff/landing fringes that the trim
thresholds (per spec) correctly include. See
_docs/03_implementation/batch_106_cycle3_report.md "Spec drift
surfaced" for the full note.
Semantics decision documented inline: max_waypoints is enforced
only in auto-tolerance mode; with an explicit DP tolerance the
result reflects that exact tolerance.
AZ-836 moved to done/.
Co-authored-by: Cursor <cursoragent@cursor.com>
Real derkachi.tlog covers 3 takeoffs at the same field but the
uploaded video covers only the last. Original NCC argmax + AZ-405
head-takeoff fallback both biased toward flight 1, violating the
spec's "the last chunk in tlog is relevant" framing.
Patch: pre-NCC flight segmenter partitions the IMU energy stream
into distinct flights (threshold + gap walk); find_aligned_window
restricts NCC search to the last segment; low-confidence fallback
uses that segment's start instead of head-takeoff detection.
AlignedWindow gains flight_count_detected + selected_flight_index
for FDR-visible audit.
7 new unit tests (segmenter shapes + end-to-end multi-flight
pipeline + segmented fallback path). 19 AZ-698 tests pass, 113
in the regression slice. Zero new mypy --strict errors.
Co-authored-by: Cursor <cursoragent@cursor.com>
Adds find_aligned_window cross-correlation (NCC, per-window unit norm)
between IMU energy and video optical-flow magnitude. Returns
AlignedWindow{tlog_start_ns, tlog_end_ns, offset_ms, confidence,
used_fallback}, with fallback to head-takeoff on low confidence to
preserve AZ-405 behavior. TlogReplayFcAdapter honors tlog_start_ns and
skips pre-window messages. New --auto-trim CLI flag, mutex with
--time-offset-ms. AC-1..AC-4 covered by unit tests; AC-5 skipped (no
real flight_derkachi.mp4 in repo). 106 tests pass in regression slice.
Zero new mypy --strict errors.
Co-authored-by: Cursor <cursoragent@cursor.com>
Batch 98 (cycle 2) — first two PBIs of epic AZ-696 (real-flight
validation harness):
AZ-697: direct binary-tlog GPS-truth extractor
- New src/gps_denied_onboard/replay_input/tlog_ground_truth.py reads
GLOBAL_POSITION_INT (with GPS_RAW_INT fallback) from a binary
ArduPilot tlog via pymavlink.mavutil and returns a frozen+slotted
TlogGroundTruth DTO with per-record ts_ns / lat_deg / lon_deg / alt_m
/ hdg_deg / vx_m_s / vy_m_s / vz_m_s.
- Promoted l2_horizontal_m + match_percentage + GroundTruthRow from
tests/e2e/replay/_helpers.py into the new production module
src/gps_denied_onboard/helpers/gps_compare.py. The e2e helper now
re-exports the same objects (identity, not copies) so existing test
imports continue working untouched.
- tests/e2e/replay/conftest.py prefers the real derkachi.tlog when
present, falls back to the CSV synth path otherwise.
- 22 new unit tests cover AC-1..AC-5 (mypy --strict subprocess test
included). All passing.
AZ-702: Topotek KHP20S30 factory-sheet camera calibration
- New _docs/00_problem/input_data/flight_derkachi/khp20s30_factory.json:
fx = fy = 4644.444, cx = 960, cy = 540, HFOV ~ 23.3 deg, VFOV ~ 13.2
deg, computed from the published 8.5 mm focal length + 1/2.8" sensor
+ 1920x1080 capture at lowest zoom step. Distortion zeroed,
body_to_camera_se3 = identity with nadir convention. Acquisition
method explicitly recorded as factory_sheet so downstream code can
expect higher residual error than a lab calibration.
- _docs/00_problem/input_data/flight_derkachi/camera_info.md updated
to document the assumptions, expected residual error window, and
conftest pick-up rule.
- tests/e2e/replay/conftest.py::_calibration_path() prefers
khp20s30_factory.json when present, falls back to adti26.json.
- 9 new unit tests cover AC-1..AC-4 (schema, intrinsics traceback,
doc reference, conftest pick-up). All passing.
Test run: 45 new tests, all passing. Full-suite gate deferred to
Step 16 (after the last batch in cycle 2 per the implement skill).
Adjacent note (not fixed in this batch, recorded in the batch report):
auto_sync.py has the same redundant pymavlink type:ignore + a few
numpy/cv2 mypy --strict issues. None on this batch's path.
Refs: _docs/03_implementation/batch_98_cycle2_report.md
Refs: _docs/02_tasks/done/AZ-697_tlog_ground_truth_extractor.md
Refs: _docs/02_tasks/done/AZ-702_khp20s30_calibration.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Mid-flight fixtures (Derkachi) and stationary-still scenarios
(FT-P-01) have no take-off spike for the IMU detector and produce
false-positive video motion onsets, so the AC-9 frame-window
validator rejects every plausible offset. Add an operator-acknowledged
opt-out: a new ReplayConfig.skip_auto_sync_validation flag that
suppresses validation, paired with a hard requirement that
time_offset_ms also be set (silent-zero guard at both schema and
adapter layers).
Wired through schema -> CLI (--skip-auto-sync) -> composition root
-> ReplayInputAdapter; Derkachi e2e fixture now passes
time_offset_ms=0 + skip_auto_sync=True by default since the synth
tlog and the video share the same t=0 anchor by construction.
5 new unit tests:
* schema gate rejects skip=True without manual offset
* schema gate accepts the legal pair
* default field value is False (default-construction safety)
* adapter constructor mirrors the schema gate
* adapter open() bypasses validate_offset_or_fail when flag is set
All 38 unit tests in test_az401 + test_az405 pass on Mac.
Co-authored-by: Cursor <cursoragent@cursor.com>
Adds the Layer-4 cross-cutting `replay_input/` module per ADR-011:
ReplayInputAdapter converges (video, tlog) into the standard
FrameSource + FcAdapter + Clock surfaces the airborne composition
root consumes. Owns time-alignment between video frames and tlog
IMU/attitude ticks (manual via --time-offset-ms or auto via the
AZ-405 IMU-take-off detector + Farneback motion-onset detector).
Auto-sync algorithm (auto_sync.py):
- Tlog take-off detector: sustained vertical-accel excess > 0.5 g for
>= 0.5 s + sustained attitude-rate magnitude > 1 rad/s.
- Video motion-onset detector: dense Farneback flow magnitude > 1.5 px
sustained >= 0.5 s (deterministic per AC-10).
- compute_offset combines the two; confidence = min(tlog, video).
- validate_offset_or_fail implements the AC-9 95 % frame-window match
validator with configurable threshold + window.
ReplayInputAdapter.open() ordering (AC-13):
1. Load tlog samples + fail-fast on missing RAW_IMU/SCALED_IMU2 or
ATTITUDE BEFORE any video read.
2. Resolve offset (auto-sync OR manual override; manual bypasses the
detectors entirely per AC-8).
3. Run AC-9 validator on resolved offset; raise auto-sync hard-fail
for AC-7 (CLI exit 2 mapping).
4. Build single Clock instance per pace (TlogDerived/ASAP, Wall/REAL).
5. Construct VideoFileFrameSource and TlogReplayFcAdapter with the
resolved offset baked in (replay protocol Invariant 8).
Structured log + FDR records on auto-sync detected / low-confidence /
AC-8 hard-fail kinds. Idempotent close (AC-12).
Tests: 25 unit tests across tests/unit/replay_input/ covering all 13
ACs (kernel-level synthetic fixtures for AC-1..AC-10; coordinator-
level OpenCV synthetic videos + faked pymavlink for AC-6..AC-13).
Contract update: replay_protocol.md v2.0.0 added fdr_client to the
ReplayInputAdapter __init__ signature (was missing in the prose; the
task spec already listed it in the allowed-imports section).
Co-authored-by: Cursor <cursoragent@cursor.com>
Oleksandr Bezdieniezhnykh