Add robust linear-Gaussian MAP smoothing

[FlorianPfaff]

Summary

• add a generic sparse robust MAP smoother for linear-Gaussian state sequences;
• support arbitrary state/measurement dimensions, transition offsets, vector-valued factors, and trusted non-robust factors;
• provide full-interval and timestamp-based fixed-lag APIs;
• expose linear, Huber, soft-L1, Cauchy, and arctangent losses;
• use one IRLS weight per whitened vector factor, preserving coordinate-rotation invariance;
• document the API and explicitly report covariances=None until MAP marginals are computed correctly.

Design

The public model is

x[k+1] = F[k] x[k] + b[k] + w[k]
z[j]   = H[j] x[k(j)] + d[j] + v[j]

Prior and process factors remain quadratic. Robust measurement factors are solved with sparse LSMR-based IRLS and a monotone backtracking line search. Fixed-lag smoothing solves the window beginning at each timestamp and returns the first state plus per-window diagnostics.

The implementation deliberately does not reuse filtered covariances as smoother covariances. Selected inverse-Hessian blocks can be added later as a separate marginal-covariance API.

Regression coverage

• linear-loss solution against dense weighted least squares;
• gross measurement outlier downweighting;
• vector measurements and deterministic offsets;
• fixed-lag window boundaries and batch parity for a covering lag;
• single-state empty transition sequences;
• malformed lag, factor index, and observation-shape rejection.

Validation

• focused local test module: 12 passed;
• production and regression modules pass py_compile;
• documentation example executed successfully.

Repository CI provides the authoritative multi-backend, lint, documentation, packaging, and integration validation.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	28.16s
✅ JSON	prettier	7	0	0	0	1.57s
✅ JSON	v8r	7		0	0	3.55s
✅ MARKDOWN	markdownlint	69	0	0	0	2.19s
✅ MARKDOWN	markdown-table-formatter	69	0	0	0	0.49s
✅ PYTHON	black	1867	88	0	0	109.9s
✅ PYTHON	isort	1867	142	0	0	3.21s
✅ REPOSITORY	betterleaks	yes		no	no	2.34s
✅ REPOSITORY	checkov	yes		no	no	63.98s
✅ REPOSITORY	gitleaks	yes		no	no	19.13s
✅ REPOSITORY	git_diff	yes		no	no	0.15s
✅ REPOSITORY	secretlint	yes		no	no	105.27s
✅ REPOSITORY	syft	yes		no	no	5.77s
✅ REPOSITORY	trivy-sbom	yes		no	no	6.98s
✅ REPOSITORY	trufflehog	yes		no	no	36.32s
✅ YAML	prettier	11	0	0	0	0.9s
✅ YAML	v8r	11		0	0	14.08s
✅ YAML	yamllint	11		0	0	0.59s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

• Documentation: Custom Flavors
• Command: npx mega-linter-runner@9.6.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository

[FlorianPfaff]

Closing as a high-maintenance, evidence-incomplete integration rather than merging a red 1,206-line solver. The package, documentation, lint, CodeQL, security, and dependency checks passed, but every backend test-matrix job failed (NumPy, JAX, and PyTorch across supported Python versions). The PR also intentionally omits MAP marginal covariances (covariances=None), while its downstream RaFT-UAV adapter is pinned to this unpublished implementation. The branch and commits remain recoverable if the smoother is later split into a smaller core PR with a fully green multi-backend matrix and a separately validated covariance/API contract.