See what `calibrex check` can check on your own bag.

Drop a ROS 2 bag and, if you have one, the calibration file your robot runs. Calibrex reads the transforms, classifies the sensor topics, and lists every sensor pair it could check, or why it could not. What the check does.

Runs locally via Pyodide: files are read in place and never leave this page. Plan the check, then run the estimators here on a short slice of the bag, or on the whole bag with the CLI.

New here? Plan a tiny synthetic bag (266 KB, an IMU, two LiDARs, a camera, a GNSS receiver and /tf_static) and see the full result in a few seconds.

Or plan your own bag

Drop a rosbag2 folder here, or its metadata.yaml and .db3 / .mcap file(s). A bare .mcap works too.

Multi-gigabyte files are fine: Calibrex reads only the pages it needs from your disk (Emscripten WORKERFS) and never copies the bag into memory. With an .mcap, add its metadata.yaml to skip a scan of the whole file.

Calibration file (optional)

The extrinsics your robot runs, if they are not already in the bag's /tf_static: a URDF, a slac.check_frames YAML, a Kalibr camchain-imucam, an RTK-SLAM calib.yaml, or a Hilti lidar_calibration.yaml. A file overrides /tf_static for the same child frame (--tf).

Vehicle frame and frame map (optional)

Vehicle frame Opt-in. The vehicle pairs (lidar-vehicle, imu-vehicle, ins-lidar, lidar-wheel_odometry) assume a ground vehicle that drives non-holonomically, so they are planned only when you name its frame (--vehicle-frame). Leave it empty for a drone, an arm, or a hand-held rig. Frame map One TOPIC=FRAME per line (--frame-map): point a topic at the sensor frame it really comes from, when its header names another frame (for example streams already transformed into base_link). Topic kinds One TOPIC=wheel or TOPIC=ins per line (--topic-kind): classify an odometry or twist topic when its name does not say (for example /oxts/twist=wheel).

Starting Python (about 30 MB the first time)…

Plan for

Candidate calibration

Candidate extrinsics are the transforms your robot deploys; calibrex check judges them against what the sensor data says.

Frame tree

Sensor pairs

The vehicle pairs are opt-in: they assume a ground vehicle. If this is one, enter its frame (for example base_link) and plan again.

Topics

Run the estimators in your browser

Runs the same pair estimators as the CLI in Pyodide, on the first part of the bag. The bag is still read in place from your disk. A browser is slower than the CLI (about 2 times on the LiDAR-odometry pairs), so keep the slice short; run the whole bag with the CLI below.

Mode Analyse the first (seconds) The duration cap (--max-duration-s). 0 reads the whole bag, which can take very long and a lot of memory in a browser tab.

Camera (optional) For the camera pairs: the image topic or frame to use (--camera), when the bag has several cameras. Empty uses the first camera of each pair.

Pairs to run

Run the whole bag on your machine

The full run reads the whole bag and can take minutes. With the CLI you also get the cache, every camera, and all the verdict options:

Replace the bag path with where it lives on your machine. Install once with:

The plan download is the slac.calibration_check/v0.1 artifact with provenance (input digest, Calibrex version, and the equivalent command); check it with calibrex validate.