docs(sessions): record the end-to-end R-IEKF EuRoC harness day - #431
Conversation
Session log + CHANGELOG milestone for #429: the R-IEKF backend driven through the validated front end on real EuRoC, measured against the standard MSCKF. R-IEKF runs stably (Joseph FullCovariance) but underperforms on ATE and position-σ; the load-bearing finding is that position-norm ATE/σ cannot test the yaw-gauge thesis — a gauge-anchored attitude-NEES is the #365 follow-up. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Pro Run ID: 📒 Files selected for processing (2)
WalkthroughThis PR adds a CHANGELOG milestone entry and a new session document recording an end-to-end R-IEKF EuRoC harness evaluation. It describes harness components, two real-data bugs found (covariance representation, non-finite update handling), measured ATE/σ results, and follow-up work. ChangesR-IEKF EuRoC Harness Session Documentation
Estimated code review effort: 1 (Trivial) | ~3 minutes Possibly related PRs
Poem 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Summary
Day wrapup for #429 — the end-to-end R-IEKF EuRoC harness.
docs/sessions/2026-06-30-riekf-euroc-harness.md— the arc, the two real-data bugs the harness surfaced (SqrtCovariance divergence → Joseph FullCovariance; non-finite-δx update guard), the measured MSCKF-vs-R-IEKF 2×2, and the load-bearing finding.## Milestones), linking the session log, markedmain, post-v0.65.0(feat(tools): drive R-IEKF through the real EuRoC pipeline #429 landed after the v0.65.0 release PR, currently unreleased).The finding, in one line
R-IEKF runs stably end to end but underperforms the body-frame MSCKF (ATE 1.3 m vs 0.27 m; position ATE/σ 17–19 vs 2.1) — and position-norm ATE/σ cannot test the yaw-gauge thesis (standard MSCKF is already ~calibrated at 2.1 on position, since the leak is rotational). A gauge-anchored attitude-NEES is the #365 follow-up.
Changes
docs/sessions/2026-06-30-riekf-euroc-harness.md— new session logCHANGELOG.md— new milestone entryDocs only — no code changes.
Relates to #365
🤖 Generated with Claude Code
Summary by CodeRabbit