Add RB-FFBSi smoother for MEM-RBPF trackers

[FlorianPfaff]

Summary

This PR adds a PyRecEst-native fixed-interval Rao-Blackwellized FFBSi smoother for MEMRBPFTracker forward passes.

It includes:

• MEMRBPFForwardRecord: detached weighted filtering records for MEM-RBPF scans.
• RBFFBSiResult: structured smoother output with point estimates, kinematic RTS results, backward-sampled particle trajectories, and sample states.
• MEMRBPFFFBSiSmoother: fixed-interval RB-FFBSi smoother.
• aliases MEMRBPF_FFBSiSmoother and RBFFBSiSmoother.
• exports in pyrecest.smoothers.
• focused tests for deterministic smoothing, stochastic shape/trajectory outputs, mapping records, aliases, and tracker snapshot construction.

Why

The MEM-RBPF tracker already has a Rao-Blackwellized filtering structure: a shared Gaussian kinematic state, orientation particles, and conditional Gaussian semi-axis states. FFBSi is a natural offline smoother for this representation.

The reproduction repository had an FFBSi smoother implemented as a local tracker subclass. This PR instead adds a reusable smoother API around explicit forward records, so applications can either collect records manually or build them from MEMRBPFTracker instances.

Main implementation choices

• The smoother consumes explicit MEMRBPFForwardRecord objects rather than owning the forward filtering loop.
• MEMRBPFForwardRecord.from_tracker(tracker) supports a future tracker.get_filtering_state(...) hook when available, and otherwise falls back to current tracker arrays.
• The kinematic state is smoothed once with a standard RTS recursion.
• Orientation trajectories are sampled backward using wrapped-normal transition densities.
• Conditional semi-axis trajectories use the linear-Gaussian backward kernel.
• Point estimates are formed from the backward samples by averaging ellipse extent matrices and eigendecomposing the mean extent.
• Output states use full axis lengths by default, matching MEMRBPFTracker.get_state(full_axis_lengths=True), but the smoother supports semi-axis output with full_axis_lengths=False.

Files changed

• src/pyrecest/smoothers/mem_rbpf_ffbsi_smoother.py
• src/pyrecest/smoothers/__init__.py
• tests/smoothers/test_mem_rbpf_ffbsi_smoother.py

Tests included

The tests cover:

• smoother aliases;
• deterministic two-record RTS and axis backward-kernel behavior;
• finite outputs for multiple particles;
• MEMRBPFForwardRecord.from_tracker(...) after a tracker update;
• mapping-record inputs.

Suggested validation

python -m pytest tests/smoothers/test_mem_rbpf_ffbsi_smoother.py

Syntax validation performed while drafting:

python -m py_compile \
  src/pyrecest/smoothers/mem_rbpf_ffbsi_smoother.py \
  tests/smoothers/test_mem_rbpf_ffbsi_smoother.py

Notes

This intentionally does not port the reproduction repository's MEMRBPF_FFBSi tracker subclass as-is. The upstream version separates recording from smoothing and exposes a reusable smoother API around explicit forward records.

The branch was created from a recent main, but main moved while this was being pushed. The changes are additive except for smoothers/__init__.py, so rebasing should be straightforward if GitHub reports the branch as behind.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	21.11s
✅ JSON	prettier	2	0	0	0	0.48s
✅ JSON	v8r	2		0	0	2.96s
✅ MARKDOWN	markdownlint	28	0	0	0	1.36s
✅ MARKDOWN	markdown-table-formatter	28	0	0	0	0.35s
✅ PYTHON	bandit	560		0	0	9.69s
✅ PYTHON	black	560	4	0	0	16.77s
✅ PYTHON	flake8	560		0	0	5.99s
✅ PYTHON	isort	560	6	0	0	1.05s
✅ PYTHON	mypy	560		0	0	7.5s
✅ PYTHON	pylint	560		0	0	101.72s
✅ PYTHON	ruff	560	6	0	0	0.11s
✅ REPOSITORY	checkov	yes		no	no	30.71s
✅ REPOSITORY	gitleaks	yes		no	no	13.93s
✅ REPOSITORY	git_diff	yes		no	no	0.04s
✅ REPOSITORY	osv-scanner	yes		no	no	0.89s
✅ REPOSITORY	secretlint	yes		no	no	9.86s
✅ REPOSITORY	syft	yes		no	no	3.53s
✅ REPOSITORY	trivy-sbom	yes		no	no	4.0s
✅ REPOSITORY	trufflehog	yes		no	no	22.63s
✅ YAML	prettier	5	0	0	0	0.53s
✅ YAML	v8r	5		0	0	5.61s
✅ YAML	yamllint	5		0	0	0.53s

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@9.5.0 --custom-flavor-setup --custom-flavor-linters PYTHON_PYLINT,PYTHON_BLACK,PYTHON_FLAKE8,PYTHON_ISORT,PYTHON_BANDIT,PYTHON_MYPY,PYTHON_RUFF,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_OSV_SCANNER,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository

[FlorianPfaff]

@copilot Fix megalinter issues

[Copilot]

@copilot Fix megalinter issues

Fixed the MegaLinter issue by updating the IDKF typing in information_form_distributed_kalman_filter.py (mypy-related) and pushed commit 3c9d3dd.