Record experiments with rosbag (MCAP)

Capture a complete telemetry trace of any bringup — joints, FSM transitions, safety status, policy I/O — to a MCAP file. MCAP is the cross-language replacement for the legacy sqlite3 rosbag backend; it's the default in ROS 2 Jazzy and what Foxglove Studio / mcap-cli consume directly.

Record everything

The one-liner most operators want:

ros2 bag record -s mcap --all -o lite_$(date +%Y%m%d_%H%M%S)

Breaking down the flags:

Flag	Meaning
`-s mcap`	Use the MCAP storage plugin. Without this you get the legacy `sqlite3` `.db3` format.
`--all`	Subscribe every topic visible at start time. Late-joiners are not picked up automatically — restart the recorder if you launch new nodes.
`-o <dir>`	Output directory name. The recorder writes `<dir>/<dir>_0.mcap` plus a `metadata.yaml`. Stamping with `$(date …)` keeps successive runs from colliding.

Stop with Ctrl+C. The MCAP file is finalized on shutdown.

Record a focused subset

--all is heavy on disk and wall-clock CPU. For a tuning session where you only care about the control loop, name the topics explicitly:

ros2 bag record -s mcap -o tuning_$(date +%Y%m%d_%H%M%S) \
    /lite/joint_states \
    /control_mode \
    /standby_controller/state \
    /safety_status \
    /joy \
    /tf /tf_static

For a policy session, also add the policy command topic /remote_policy_controller/command (bar_msgs/MITCommand). For piano runs, capture /piano/target_keys and /piano/key_state; both are pianist_msgs/PianoKeyState, and the pair is the source for offline F1 / precision / recall computed via pianist_metrics.

Tips

Where to put bags: pick a path outside the workspace (~/bags/...) so colcon build doesn't try to index multi-GB .mcap files. The workspace's .gitignore already excludes bag_* / *.mcap at the repo root, but a path under ~ is safest.
TRANSIENT_LOCAL topics (e.g. /robot_description, /safety_status): the recorder captures the latched value the moment it subscribes, so start ros2 bag record after the launch is up if you want the URDF in the bag.
Sim time: when recording from MuJoCo the timestamps come from /clock, so the bag is already on sim time. Replay with --clock and start downstream consumers with use_sim_time:=true if you want consistent now() semantics across the playback.
Compression: MCAP supports built-in zstd compression — add --compression-mode message --compression-format zstd. Roughly 2× shrink on typical float-heavy /lite/joint_states traffic, at ~5% CPU overhead.

Inspect a recording

# Topic list + message counts + duration
ros2 bag info -s mcap <bag-dir>

# Or use mcap-cli directly on the file (no ROS needed):
mcap info <bag-dir>/<bag-dir>_0.mcap
mcap list channels <bag-dir>/<bag-dir>_0.mcap

mcap-cli ships in the pixi env (available as mcap … inside pixi shell). For interactive plotting, drop the .mcap into Foxglove Studio (foxglove-studio package, also pixi-installed).

Replay

ros2 bag play -s mcap <bag-dir>

For replaying into a stack that uses sim time, add --clock 200 (publishes /clock at 200 Hz) and launch your consumers with use_sim_time:=true. Note that replay does not re-run the controllers — it just re-publishes the recorded topics. To replay an action stream into the live controller chain, see Switch controllers manually.

Post-process piano runs

The pianist_metrics numpy package (installed via pixi pypi-deps from the upstream pianist-tracking-mj repo) is the source of truth for piano evaluation metrics. Capture both topics during the run:

ros2 bag record -s mcap -o piano_$(date +%Y%m%d_%H%M%S) \
    /piano/target_keys /piano/key_state

Then offline:

import pianist_metrics as pm
# target, pressed: (T, 88) bool arrays from the bag
precision = pm.precision(target, pressed)
recall    = pm.recall(target, pressed)
f1        = pm.f1(target, pressed)

There is no live ROS node computing these — bag-based post-processing is the deployment-side evaluation path.

Record experiments with rosbag (MCAP)

Record everything​

Record a focused subset​

Tips​

Inspect a recording​

Replay​

Post-process piano runs​

See also​