Repository navigation
A real front end's errors and the stochastic clone (WB-3); ADR-0023 - #80
Merged
Merged
Conversation
…nd the error study that uses them Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
…clone against them; ADR-0023 Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Comment on lines
+342
to
+343
| "| Variant | Runs | Median mean NEES | Median 2-sigma coverage | Median final error m | Median claimed sigma m " | ||
| "| Meets the gate (NEES < 10, coverage > 90%) |", |
Comment on lines
+386
to
+388
| "| Gap, frames | Pairs | Median baseline m | Rotation RMS deg | Lateral RMS m | Median direction error deg " | ||
| "| Outlier share | Rotation lag-1 autocorrelation | Lateral lag-1 autocorrelation " | ||
| "| Rotation correlation time s | Lateral correlation time s |", |
| direction = [by[g]["dir_median_deg"] for g in (1, 2, 4, 6, 10)] | ||
| outliers = [by[g]["outlier_share"] for g in (1, 2, 4, 6, 10)] | ||
| lateral = [by[g]["lat_rms_m"] for g in (1, 2, 4, 6, 10)] | ||
| assert direction == sorted(direction, reverse=True) and direction[0] > 10 > 2 > direction[-1] - 0.0 |
| R = rot_exp(np.array([0.01, -0.02, 0.03])) | ||
| u = np.array([0.6, 0.0, 0.8]) | ||
|
|
||
| class Truth: |
…n the lag window Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
…cannot delete an open file) Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
navkit.vo_frontend: a NumPy-only two-view front end (Shi-Tomasi, pyramidal Lucas-Kanade with a forward-backward check, eight-point RANSAC, cheirality), chained across frames.navkit euroc fetch --images: fetches left-camera frames of a window of one sequence, from a stored, compressed or whole local archive.navkit euroc vo: measures the front end's rotation and translation error against the ground truth, outliers and correlation time, and checks the stochastic clone against the measured errors (a Gaussian surrogate and a replay of the real error series).docs/real_front_end.md, committed CSVs, tests that check each statement against them, and ADR-0023 (accepted).Result (EuRoC MH_01_easy, 30 s, 20 Hz)
Limits
One sequence, one 30 s window, and one replayed error series (the seeds change only the simulated GNSS, so 91% has no spread behind it). The front end is mine and plain, and the translation has the true scale, which a monocular front end lacks. The default is unchanged; the clone stays opt-in.
Full suite (1298 passed), lint, types, links, strict docs build, SRS trace and metrics pass locally. Coverage 94.34%. The camera frames are not in the repository (git-ignored, non-commercial terms); the tests use synthetic frames and the committed CSVs.
🤖 Generated with Claude Code