Run MOLA on your own ROS 2 bag

You have a bag with a 3D LiDAR in it and you want a trajectory and a map out of it. This page answers the five questions you will hit, in the order you hit them.

The short version, if your bag is well behaved:

MOLA_LIDAR_TOPIC=/your/points \
  mola-lo-gui-rosbag2 /path/to/your_bag.mcap

A directory containing metadata.yaml works in place of the .mcap file.

1. Which topic is my LiDAR?

ros2 bag info /path/to/your_bag.mcap

Find the topic whose type is sensor_msgs/msg/PointCloud2, and pass it as MOLA_LIDAR_TOPIC. The default is /ouster/points, which is almost certainly not what your bag calls it.

Set MOLA_IMU_TOPIC too if the bag carries a sensor_msgs/msg/Imu. It is optional, but an IMU improves deskewing and, on some rigs, is what makes the run work at all.

Note

MOLA needs per-point timestamps to deskew a spinning LiDAR properly. Most drivers publish them, but not all, and a cloud without them is processed as if every point were captured at the same instant. On a vehicle at road speed that is a visible distortion, not a rounding error.

2. Do I have `/tf`?

This decides whether MOLA tracks your vehicle or your sensor.

If the bag carries the transform from base_link to the LiDAR frame, on either /tf_static or /tf, MOLA uses it and the trajectory is your vehicle’s. Nothing to configure; a dynamically published transform is fine.

If it does not, which is common when only a driver node was recorded, tell MOLA where the sensor sits:

MOLA_USE_FIXED_LIDAR_POSE=true \
LIDAR_POSE_X=0.0 LIDAR_POSE_Y=0.0 LIDAR_POSE_Z=1.6 \
LIDAR_POSE_YAW=0 LIDAR_POSE_PITCH=0 LIDAR_POSE_ROLL=0 \
MOLA_LIDAR_TOPIC=/your/points \
  mola-lo-gui-rosbag2 /path/to/your_bag.mcap

Positions are in meters, angles in degrees. Leaving all six at zero is a valid choice: it means “track the sensor, and I know it”.

For a namespaced bag (/robot1/tf), also set MOLA_TF_TOPIC and MOLA_TF_STATIC_TOPIC, and MOLA_TF_BASE_LINK if your base frame is not called base_link.

Note

Those two topic overrides need MOLA 2.7.0 or newer. Before that, mola_input_rosbag2 read /tf and /tf_static by hard-coded name, so on an older install a namespaced bag is silently ignored rather than rejected.

3. GUI or CLI?

Use mola-lo-gui-rosbag2 the first few times: you want to watch the cloud and see whether the trajectory looks sane.

Switch to mola-lo-cli-rosbag2 for anything you intend to measure. The GUI drops scans under real-time pacing and is not reproducible. This matters more than it sounds and is worth reading GUI or offline CLI: which one, and why before you publish a number.

4. Which pipeline?

Do not choose one yet. The default is tuned to be reasonable across very different sensors and environments, and it is the right starting point for a bag nobody has seen before.

Change it only once you have a run to compare against, and see LO/LIO pipelines for what the alternatives actually do. The per-dataset pages are worth reading here too: each records which pipeline that dataset needed and the measurements behind the choice, which is a better guide than picking one by name.

5. Where did my map go?

By default, nowhere. MOLA-LO computes a trajectory; saving a map is opt-in:

MOLA_GENERATE_SIMPLEMAP=true \
MOLA_SIMPLEMAP_OUTPUT=my_map.simplemap \
MOLA_LIDAR_TOPIC=/your/points \
  mola-lo-cli-rosbag2 /path/to/your_bag.mcap

A simple-map is the keyframe poses plus their raw observations. It is the input to map building, not the finished map: turning it into a metric map you can localize against is covered in Tutorial: build a map.

You can also toggle this from the mola_lidar_odometry sub-window while the GUI is running.

When it does not work

See Troubleshooting. The failure modes are few and mostly about /tf, topic names and timestamps.

For the full matrix of live-node, namespaced, geo-referenced and external-odometry setups, see ROS 2 configurations.