Sensors and datasets

This page lists the ways to feed sensory data into a MOLA system, offline from a recording or online from a live sensor.

Every option below implements the same RawDataSourceBase interface, so the choice of input is independent of the choice of odometry, SLAM or localization algorithm: the same pipeline runs unchanged on a rosbag, on a rawlog, or on a live LiDAR.

1. Offline: from a recording

Generic formats

Format	Reader module	Run LiDAR odometry on it with
ROS 2 bag (`.mcap`, sqlite3)	mola_input_rosbag2	`mola-lo-gui-rosbag2 <bag>` (GUI) or `mola-lo-cli-rosbag2 <bag>` (offline)
ROS 1 bag (`.bag`)	mola_input_rosbag1, which needs no ROS 1 installation	`mola-lo-gui-rosbag1 <bag> [more bags]` or `mola-lo-cli-rosbag1 <bag> [more bags]`
MRPT rawlog (`.rawlog`)	mola_input_rawlog	`mola-lo-gui-rawlog <file.rawlog>` or `mola-lo-cli-rawlog <file.rawlog>`
KITTI-style ``.bin`` directories	mola_input_lidar_bin_dataset	No odometry launcher yet; to replay one, use `mola-cli $(ros2 pkg prefix mola_demos)/share/mola_demos/mola-cli-launchs/lidar_bin_directory_just_replay.yaml`

Note

The mola-lo-* launchers are installed by mola_lidar_odometry. Each is a thin wrapper that picks the right mola-cli launch file (GUI) or supplies the right flags to mola-lidar-odometry-cli (offline), so you do not have to pass -c <pipeline.yaml> and --state-estimator-param-file yourself. Run any of them with no arguments to see its usage, and see Launching MOLA-LO for the underlying commands.

Note

Several bags can be replayed jointly as one merged stream, which is what datasets publishing one bag per topic require. See the LO launchers.

To convert between these formats, see dataset conversions.

Public benchmark datasets

MOLA ships a ready-to-run launcher per dataset, so you do not have to work out its topic names, frame ids, extrinsics or per-dataset tuning yourself. Each comes in two flavors: mola-lo-gui-<name> for interactive replay with the 3D GUI, and mola-lo-cli-<name> for offline batch processing at full speed.

The table below is the overview. For one dataset in particular, see its own page: where to put the data, the exact commands, and every tuned default with the reasoning behind it.

Dataset	What it is	Launcher suffix
KITTI	The odometry benchmark; a car with a Velodyne HDL-64.	`kitti`
KITTI-360	Panoramic successor to KITTI.	`kitti360`
MulRan	A car with an Ouster OS1-64 around Daejeon and Sejong.	`mulran`
Newer College	Handheld rig walked around New College, Oxford; ships a survey-grade prior map.	`newer-college`
Oxford Spires	Hand-held / backpack VILENS payload around Oxford colleges.	`oxford-spires`
Hilti-Oxford 2022	Hand-held survey pole through construction sites and the Sheldonian Theatre.	`hilti2022`
GrandTour	ANYbotics ANYmal-D quadruped carrying the “Boxi” payload; one bag per topic.	`grandtour`
BotanicGarden	Ground robot driven through a botanical garden.	`botanicgarden`
CitrusFarm	Clearpath Jackal through citrus orchard rows.	`citrusfarm`
ConSLAM	Hand-held scanner in construction sites (ROS 1 or ROS 2 bags).	`conslam`
GEODE	Built around degenerate geometry: flat surfaces, stairwells, metro tunnels, off-road, inland waterways.	`geode`
TIERS	Five LiDARs recording simultaneously, hence one launcher per sensor.	`tiers-ouster-os0`, `tiers-ouster-os1`, `tiers-velodyne`, `tiers-livox-horizon`, `tiers-livox-avia`
Paris-LuCo	The ParisLuco128 sequence used by CT-ICP and others.	`paris-luco`

So, for example:

# Interactive replay with the GUI:
mola-lo-gui-mulran KAIST01

# The same sequence, offline and as fast as the machine allows:
mola-lo-cli-mulran KAIST01 --output-tum-path mulran-kaist01.tum

Every launcher prints its own usage, including the dataset-specific environment variables it honors, when run with no arguments.

2. Online: from a live sensor

Path	When to use it
ROS 2 topics, via mola_bridge_ros2	The usual choice: your sensor already has a ROS 2 driver publishing `sensor_msgs/PointCloud2`, `sensor_msgs/Imu`, `/tf`, … See the ROS 2 node and the ROS 2 configurations cookbook.
Ouster native SDK, via mola_input_ouster	Direct connection to an Ouster sensor (or replay of a PCAP/OSF recording) with no ROS middleware in the loop. Launcher: `mola-lo-gui-ouster`.
mrpt-hwdrivers	Direct connection to sensors supported by MRPT without ROS: 2D lidars (Hokuyo, SICK, RP-Lidar, Ibeo), Velodyne 3D lidars, cameras (OpenCV / ffmpeg / Bumblebee2), depth sensors (Kinect, OpenNI2), IMUs (xSens, KVH DSP3000), and NMEA or Novatel GNSS receivers.

3. What if my sensor or dataset is not listed?

Nothing above is a special case in the odometry code. If your data is in a ROS 1 or ROS 2 bag, or in a rawlog, MOLA can already read it: what the per-dataset launchers add is only the knowledge of which topic is which. Point the generic launcher at your recording and name the topics:

MOLA_LIDAR_TOPIC=/ouster/points \
MOLA_IMU_TOPIC=/ouster/imu \
mola-lo-gui-rosbag2 /path/to/your/dataset.mcap

See the rosbag2 launcher for the full list of options, including what to do when your bag has no /tf.