Roadmap

Intentions, not commitments with dates.

Documentation

Item	Status
Theory pages — full derivations and Jacobians	in progress; ported from camera/CameraModel, the unified factor, and the ACI / t_ga derivation
Generated CLI reference	planned — CLI is hand-written and can drift
Generated configuration reference	planned — Configuration is hand-written and can drift
Generated Python API from docstrings	undecided (O4) — Python API is a hand-written map today
Simulation guide — authored trajectories	in progress; today’s simulation is digital-twin only

Toolbox

Item	Status
Continuous-time refinement	planned — see Continuous-time refinement
Authored-trajectory synthetic generator	planned — would let degeneracy be studied without recording restricted motion
API stability guarantee	not yet — the CLI and config schema are the stable surface
omni_radtan out of experimental	in progress — already validated on TUM-VI camera–IMU

Datasets

Item	Status
Additional example datasets	open — TUM VI, Monado, and Looper SA16 are complete
TUM VIE	needs data — not acquired, no runs; the natural successor to TUM VI
Dataset hosting on Hugging Face with Zenodo DOIs	in progress
Contributed datasets	open — Submit a dataset

Benchmark

Item	Status
Submission CI validation	in progress
Maintainer re-run / verified badge	planned — Governance
Versioned, citable leaderboard snapshots	planned
Annual calibration challenge	under consideration

Site

Item	Status
LICENSE file in the toolbox repo	needed — MIT is declared in pyproject.toml only, see License
Public toolbox repo	pending — CamCalib is private, so the header GitHub link points at the site repo
Real logo replacing the placeholder	planned
Versioned documentation per release	undecided — single “latest” today
Release tags and dated changelog entries	planned — see Changelog

Not planned

• Replacing Kalibr for single-IMU rigs. It is mature and widely validated; CamCalib exists for the multi-IMU, multi-model, no-ROS case. See Comparison.
• A composite leaderboard score. Weighting rotation error against runtime is the reader’s judgement, not the benchmark’s — see Methodology.

Something missing that you need? Open an issue on the repository.