Looper SA16

The looperrobotics SA16 rig is the example closest to production use: a commercial stereo (later three-camera) module recorded four different ways over several months. It is the dataset where CamCalib was pushed past a single clean sequence into rolling shutter, IMU filter changes, and irregular timestamps — and where cross-campaign agreement, not a single residual, is the evidence.

The rig

Sensor	Rate	Detail
cam0, cam1	20 Hz	544×640 grayscale, hardware-synced
cam2 (SA16 7.3 only)	4.29 Hz	rolling shutter, tr estimated
IMU	400 Hz	timestamp, omega_xyz, alpha_xyz

Board: AprilGrid 6×6, tag_size 0.055 m, tag_spacing 0.3, family t36h11.

Getting the data into shape

The recordings are not EuRoC, but they are close — one prep step reshapes them into a mav0/ tree, and no CamCalib code changes were needed for any campaign:

python examples/configs/looperrobotics/build_mav0.py \
  --root '/data/looperrobotics/SA16(96s record)'

Symlinks images, writes data.csv, comments the IMU header. The IMU CSV column order is already EuRoC-native.

The two-stage recipe

# 1. Camera-only — intrinsics + stereo extrinsics
camcalib cam -c examples/configs/looperrobotics/cam_only.yaml

# 2. Camera-IMU — T_CtoI, td, IMU intrinsics and biases, intrinsics pinned
camcalib cam-imu -c examples/configs/looperrobotics/cam_imu.yaml

Cam-IMU uses window_mode: motion, which auto-picks the most-exciting 80 s of the 96.6 s log.

Fit cameras individually, then merge

The 96 s campaign’s lesson: on this rig, per-camera fits merged afterwards beat a joint stereo fit. merge_cam_only.py combines them.

Four campaigns

SA16 (96 s) — the baseline

Stereo, radtan first then equidistant. Baseline 98.42 mm.

SA16 (7.3) — a rolling-shutter third camera

Adds cam2 at 4.29 Hz with rolling shutter and free tr.

Result	Value
cam2 readout tr	33.3 ms — 30.6 µs/line × 1088 rows
tr model-dependence	equi 33.30 ms vs radtan 33.77 ms — essentially none
cam2 reprojection	equidistant 2.02 px vs radtan 7.57 px
cam0 / cam1 reprojection	0.78 / 0.76 px
Baseline	98.70 mm (96 s: 98.42)

Un-rotate the rolling-shutter camera at prep time

cam2’s files are stored 90° rotated relative to the sensor. The rolling-shutter model sweeps image rows, so on rotated images tr collapses to nothing. The prep step fixes it:

python examples/configs/looperrobotics/build_mav0.py \
  --root '/data/looperrobotics/SA16(7.3)' --rotate cam2=cw

The recovered 33.3 ms decomposes cleanly into 30.6 µs/line × 1088 rows, which is what tells you the number is physical rather than a fitted artefact.

Equidistant beating radtan by 3.7× on reprojection is the recurring result on this rig — equidistant is canonical here.

SA16 (OSR4) — IMU oversampling ratio 4

Two ~51 s stereo recordings with the IMU’s oversampling ratio changed to 4.

Result	Value
Reprojection	0.215 / 0.216 px — the cleanest looper solves
Baseline	99.24 / 99.26 mm
t_ga	collapsed −5.19 → −1.0 ms
Camera td	shifted +0.4 ms common-mode

Geometry unchanged, timing shifted. The OSR4 filter changes the IMU’s internal group delays — differently for gyro and accel — and t_ga plus td absorb exactly that. The t_ga value was confirmed estimator-free by studies/tga_correlation_lever.py (−0.9 to −3.0 ms across bands; 20–80 Hz agrees with the solver to ~0.1 ms), so it is a measurement of the hardware, not an artefact of the graph.

SA16 (data-sync) — real hardware timestamps

Two recordings (41.7 / 44.5 s) with the IMU hardware-synced to the camera trigger. The first looper set with non-idealised IMU timestamps: ~68 µs jitter, 114/107 dropped samples, and a timebase running 766/783 ppm slow against the camera — independently observable for the first time, because every earlier set was quantised to exactly 2.5000 ms and so had a ratio of 20.000000 by construction.

Result	Value
Reprojection	0.221 / 0.210 px — ties OSR4, on more excited data
t_ga	−1.536 / −1.511 ms
Resampling needed	none
Code changes needed	none

The irregular stream costs nothing. The jittery, gappy, drifting timestamps are consumed natively. A forced-400 Hz control proves the recorded stamps are the right ones: imposing the old regular convention costs 7× in reprojection and drags td to −15.8 ms.

Round-trip

Every campaign is validated end to end. The mav0 tree is converted to a bag first, since apply-calib operates on ROS1 bags:

python examples/experiments/convert_mav0_to_bag.py \
  --mav0 '/data/looperrobotics/SA16(7.3)/mav0' \
  --out  '/data/looperrobotics/SA16(7.3)/sa16_73.bag'

camcalib roundtrip -c examples/configs/looperrobotics/sa16_73_cam_imu_ros1.yaml \
  --out results/looperrobotics/sa16_73/roundtrip --keep-corrected

What the campaigns together show

Baselines of 98.42, 98.70, 99.24, and 99.26 mm across four recording setups — different camera counts, IMU filters, and timestamp regimes — with the last two matching the robot-arm campaign. That agreement across independent recordings is a stronger accuracy claim than any single sub-pixel residual, and it is exactly what the batch sweep workflow is built to produce.

One honest caveat

The OSR4 campaign hit repeated crashes — NumPy nditer errors, corrupted int constants, GTSAM shape mismatches — traced to machine-level instability on the recording workstation, not to CamCalib. Data, code, and environment were each ruled out; the stock system Python and Kalibr crashed on the same box. Retries converged and the converged results are cross-consistent, but the episode is recorded rather than tidied away.

Reproducibility

Item	Value
Configs	examples/configs/looperrobotics/ — cam_only.yaml, cam_imu.yaml, sa16_73_*, osr4/, data_sync/
Prep	build_mav0.py (--rotate cam2=cw for SA16 7.3)
Merge	merge_cam_only.py
Board	AprilGrid 6×6, 0.055 m, spacing 0.3, t36h11
Canonical model	equidistant
Results	results/looperrobotics/{sa16,sa16_73,osr4,data_sync}/