feat(lab): publish fail-closed TGS evidence
This commit is contained in:
DCCONSTRUCTIONS
@@ -1,7 +1,8 @@
|
||||
# M4.9 gravity-aligned TGS fail-closed evidence — 2026-08-26
|
||||
|
||||
Status: **evidence adapter completed**; visual obstacle/traversability quality,
|
||||
full-source realtime, navigation and actuation remain disabled
|
||||
Status: **evidence adapter and canonical LAB publication completed**; visual
|
||||
obstacle/traversability quality, full-source realtime, navigation and actuation
|
||||
remain disabled
|
||||
|
||||
## Decision
|
||||
|
||||
@@ -13,7 +14,9 @@ accounts for every point inside TRAVEL's declared `1–80 m` processing range.
|
||||
This closes the representation and accounting gate. It does not accept visual
|
||||
quality. Several anchors still contain a high non-ground share, so the next
|
||||
gate is a 3D/costmap review of the sealed output rather than threshold tuning or
|
||||
camera boxes.
|
||||
camera boxes. The metric output is now available in the canonical LAB on port
|
||||
`8000` as an operator-review surface; publication does not upgrade it to an
|
||||
accepted terrain provider.
|
||||
|
||||
## Frozen identity
|
||||
|
||||
@@ -125,15 +128,73 @@ Worker root:
|
||||
| `inputs/input-manifest.json` | `13,357` | `7a22e61606e1fa9ab61d47417f941b590a68b29e4529d4c290992ce7eb89a642` |
|
||||
| `tgs-timing.tsv` | `638` | `a8f439996205286335ac392b37f51b54056d04e9cb07e03648c52c1e8f2f6ae8` |
|
||||
|
||||
## Canonical LAB publication
|
||||
|
||||
- LAB result:
|
||||
`m49-tgs-fail-closed-9d5cb089bb5cc23f829acb47eda52eaa886db3b60f8fa7d57e02e27f642e837b`;
|
||||
- linked current visual timeline:
|
||||
`m4-threat-replay-2a953c5f27f2a5b1dddc5c658c1de2c323d7796084a099c024987a1da03aa324`;
|
||||
- source LiDAR pack SHA-256 remains
|
||||
`0685d24219d8236caf8b7f1685e93f6d6b59e7fd015a768d88a92bbe8b154944`;
|
||||
- the earlier immutable seal beginning `200b57cc` is superseded only because it
|
||||
linked the retired M4 v2 visual profile, whose timeline is incompatible with
|
||||
the current v3 API. No Worker evidence byte or TGS result was changed.
|
||||
|
||||
The LAB reuses the canonical recorded-realtime camera and spatial instrument.
|
||||
It exposes the ten anchors, the full classified TGS point set, the four-state
|
||||
costmap, `VIDEO/CAMERA`, `3D/PLAN`, independent point/costmap visibility and
|
||||
exact per-anchor counts. Camera proposal boxes, AOS clusters, future frames and
|
||||
navigation authority are absent. Browser acceptance confirmed that every
|
||||
operator control is independent and that a loading anchor cannot temporarily
|
||||
display unrelated M4 obstacle statistics.
|
||||
|
||||
The published TGS coordinates are intentionally sensor-origin
|
||||
`map-gravity-local`: the ground is therefore approximately `-1.25 m` below the
|
||||
current LiDAR. The canonical LAB presentation converts these immutable values
|
||||
to the scene's body-ground origin by adding the rig's sealed
|
||||
`nominal_sensor_height_m = 1.25` after the gravity-to-body rotation. The same
|
||||
offset is applied to point heights and costmap z-bounds; x/y evidence and the
|
||||
sealed Worker result remain unchanged. A regression check across all ten
|
||||
anchors puts the median displayed ground points in `-0.299…+0.135 m` and the
|
||||
median ground-cell centres in `-0.308…+0.237 m`. On anchor `1856`, where the
|
||||
presentation defect was reported, those medians are `+0.004 m` and `-0.063 m`.
|
||||
The residual variation is measured terrain/pose variation, not the previous
|
||||
systematic `1.25 m` lift.
|
||||
|
||||
## Preliminary corridor triage
|
||||
|
||||
The published evidence was also projected into the frozen virtual body
|
||||
corridor (`-0.5…8.0 m` longitudinal, `±0.5 m` lateral) without changing the
|
||||
sealed result. The table counts costmap cell centres inside that corridor:
|
||||
|
||||
| Anchor | Current occupied | Rolling occupied | Rolling unobserved | Preliminary reading |
|
||||
| ---: | ---: | ---: | ---: | --- |
|
||||
| `171` | `0` | `8` | `6` | unresolved rolling band at `5.63…7.90 m` |
|
||||
| `306` | `0` | `0` | `12` | no corridor obstruction |
|
||||
| `368` | `1` | `1` | `16` | compact occupied support at `6.34 m` |
|
||||
| `402` | `0` | `1` | `10` | compact occupied support at `1.89 m` |
|
||||
| `450` | `0` | `0` | `8` | no corridor obstruction |
|
||||
| `509` | `0` | `0` | `7` | no corridor obstruction |
|
||||
| `525` | `0` | `0` | `9` | no corridor obstruction |
|
||||
| `744` | `0` | `3` | `25` | weakly observed; requires explicit review |
|
||||
| `1122` | `0` | `2` | `24` | rolling recovers compact obstacle support |
|
||||
| `1856` | `1` | `1` | `13` | compact occupied support at `4.14 m` |
|
||||
|
||||
This is a diagnostic, not ground truth. It shows why the causal window cannot
|
||||
simply be removed: anchor `1122` is entirely unobserved in the current-only
|
||||
corridor but gains ground and occupied evidence in rolling. It also identifies
|
||||
the opposing risk: at anchor `171`, rolling adds eight occupied cells where the
|
||||
camera/PLAN comparison appears to show an open route apart from side geometry.
|
||||
Anchor `744` remains too weakly observed for acceptance. These two anchors keep
|
||||
the visual gate open even though the remaining frames do not show a continuous
|
||||
occupied carpet.
|
||||
|
||||
## Next gate
|
||||
|
||||
Import the sealed `evidence.npz` into one laboratory review surface with:
|
||||
|
||||
1. 3D points colored by the four evidence states;
|
||||
2. top-down `0.45 m` costmap cells in the same local coordinate frame;
|
||||
3. anchor selection and exact counts from `result.json`;
|
||||
4. no AOS clusters, camera boxes, future frames or physical authority.
|
||||
|
||||
The operator review must focus on the mandatory obstacle/gap cases and on the
|
||||
large non-ground populations listed above. Camera projection follows only if
|
||||
the metric evidence is accepted.
|
||||
Review the published ten anchors in both metric `3D` and `PLAN`. The operator
|
||||
must focus on the mandatory obstacle/gap cases and the large non-ground
|
||||
populations listed above. Accept TGS for a full recorded-source shadow only if
|
||||
red non-ground evidence remains on real geometry without carpeting the visible
|
||||
route, soft traversable vegetation or usable gaps. Otherwise reject this
|
||||
candidate without tuning thresholds against these ten anchors. Camera
|
||||
projection follows only if the metric evidence is accepted.
|
||||
|
||||
@@ -0,0 +1,36 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Publish the verified Worker TGS pack as immutable Mission Core LAB evidence."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
from pathlib import Path
|
||||
|
||||
from k1link.laboratory.m49_tgs_fail_closed import (
|
||||
M49TgsFailClosedError,
|
||||
seal_m49_tgs_fail_closed,
|
||||
)
|
||||
|
||||
|
||||
def main() -> int:
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument("--source-root", type=Path, required=True)
|
||||
parser.add_argument("--destination-root", type=Path, required=True)
|
||||
parser.add_argument("--profile", type=Path, required=True)
|
||||
parser.add_argument("--linked-visual-result-id", required=True)
|
||||
arguments = parser.parse_args()
|
||||
try:
|
||||
result = seal_m49_tgs_fail_closed(
|
||||
source_root=arguments.source_root,
|
||||
destination_root=arguments.destination_root,
|
||||
profile_path=arguments.profile,
|
||||
linked_visual_result_id=arguments.linked_visual_result_id,
|
||||
)
|
||||
except (M49TgsFailClosedError, OSError, ValueError) as exc:
|
||||
parser.error(str(exc))
|
||||
print(result.result_id)
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
Reference in New Issue
Block a user