feat(cameracalibrate): charuco boards and a --check mode

[AaryanAgrawal]

feat(cameracalibrate): charuco boards and a --check mode

Problem

dimos cameracalibrate handled a chessboard only, and there was no way to tell whether a shipped calibration had drifted. Charuco boards recover intrinsics from partial/occluded views a plain chessboard can't, and a drifted calibration silently poisons every downstream pose estimate.

What this adds

• Charuco board support — calibrate from a charuco board, not just a chessboard. Robust to occlusion and partial views (each marker is uniquely identified). Feeds the existing cv2.calibrateCamera / cv2.fisheye.calibrate solvers.
• --check mode — re-score an existing calibration against fresh frames: per-view solvePnP reprojection RMS + drift vs a fresh calibration in the deployed lens model → an OK / DEGRADED verdict and a JSON report.
• Light coverage overlay — during capture, a hint overlay that clears where the board has already swept (space to capture, q to quit). Hint only; never gates capture.

Additive: chessboard detection, both calibration solvers, write_camera_info_yaml, and the capture loop are untouched.

Breaking Changes

---

• Board size is given as inner corners exactly (no square-count / rotated auto-guess).
• --check on a charuco target solves via cv2.calibrateCamera over the id-selected correspondences (same math as calibrateCameraCharuco).

---

Core changes — why

• utils/cli/cameracalibrate/cameracalibrate.py — one detector path per board, one cv2.calibrateCamera / cv2.fisheye.calibrate over per-view correspondences, write_camera_info_yaml for the saved intrinsics. Uses verified OpenCV throughout; is_fisheye_model is imported from marker_pose, not re-implemented.
• robot/cli/dimos.py — the cameracalibrate command wiring (flags → run_calibration / run_check_report).

How to Test

Hardware. Point a camera at a charuco (or chessboard), capture ~15 views, calibrate, then --check the result against fresh frames:

uv run dimos cameracalibrate --source webcam --board charuco --cols 9 --rows 6
uv run dimos cameracalibrate --check --calibration <cam>.yaml --source webcam

Test to read: test_cameracalibrate.py::test_write_camera_info_yaml_round_trips_through_dimos_camera_info — the saved YAML loads back identically through DimosCameraInfo. Calibration math itself is verified OpenCV.

AI assistance

Claude Code (Opus 4.8 / Fable 5) — implementation + simplification; all changes reviewed.

Checklist

• Scoped to one problem.
• Ran uv run pytest + pre-commit for the changed files.
• Reviewed every line.
• Disclosed AI assistance.
• Read + approved the CLA.