MOLA-LIO: Hand-held LiDAR mapping and localization

This hands-on tutorial will show you how to build a 3D map live using MOLA-LIO from a hand-held or robot-mounted LiDAR with an internal IMU (Ouster, Livox,…), how to post-process the saved map, and how to load that map in LIO localization-only mode.

The tutorial is fully available in this YouTube video, but you also have all the commands and steps described in text below for your convenience:

Prerequisites: MOLA installation

This tutorial assumes basic knowledge of ROS 2, networking, and Linux.

You must first install the packages mola_lidar_odometry and mola. Following the default installation instructions is enough.

1. Launching the LiDAR ROS driver

In the following we will assume an Ouster sensor, so change these commands accordingly for other manufacturers.

First, clone the ROS 2 “driver” package for your LiDAR into your ROS 2 workspace and build it:

# Clone:
cd ~/ros2_ws/src
git clone https://github.com/ouster-lidar/ouster-ros.git --recursive

# Build it:
cd ~/ros2_ws
colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=RelWithDebInfo

# And activate the new packages:
. install/setup.bash

Before going on, we need to power up the LiDAR and connect its Ethernet network cable to the computer, using a link-only Network configuration, and find the LiDAR network address (IP or local address). See the video above for step by step details.

Next, launch the “driver” to start capturing LiDAR scans. The easiest way is using the launch file ouster_ros/sensor.launch.xml which shows the live data visually in RViz (it may take a few seconds to start). A starting point command would be (customize as needed):

# Show all available options:
ros2 launch ouster_ros sensor.launch.xml --show-args

# Launch with all defaults, using computer-stamped timestamps:
ros2 launch ouster_ros sensor.launch.xml sensor_hostname:=os-XXXXXXXXXXXX.local \
  timestamp_mode:=TIME_FROM_ROS_TIME

2. Record your dataset

Now, there are two options: * Either we build the map using the live sensor, or * We record the dataset in a rosbag, then build it offline.

I would recommend grabing the data for two reasons. First, in this way, you can save several bags, browse them (using FoxGlove) and pick the one that looks best. Second, the computer load would be reduced during grabing if not running mapping live, reducing the probability of missing some LiDAR network packets.

# Show all available options:
ros2 bag record  /ouster/imu /ouster/points /tf_static

Press CTRL+C to end the recording.

Note

If you want to run live mapping instead, please check the LIO ROS 2 node documentation.

3. Build the map

Just invoke this command to launch the LIO process on your Ouster dataset (check the video for explanations on the different arguments):

MOLA_GENERATE_SIMPLEMAP=true \
MOLA_SIMPLEMAP_OUTPUT="myMap.simplemap" \
MOLA_SIMPLEMAP_MIN_XYZ=0.2 \
MOLA_LO_INITIAL_LOCALIZATION_METHOD="InitLocalization::PitchAndRollFromIMU" \
MOLA_DESKEW_METHOD=MotionCompensationMethod::IMU \
MOLA_IMU_TOPIC="/ouster/imu" \
MOLA_LIDAR_TOPIC="/ouster/points" \
MOLA_TF_BASE_LINK="os_sensor" \
mola-lo-gui-rosbag2 \
  /path/to/your/dataset.mcap

As a result, we have a reconstructed keyframe-based map in .simplemap format. We can apply any pipeline we want to build a metric map (an actual aggregated point cloud, or a voxel map, etc.). For example, let’s run:

sm2mm -i myMap.simplemap -o myMap.mm -p sm2mm_no_decim_imu_mls_keyframe_map.yaml

Inspect the resulting map with:

mm-viewer myMap.mm  -l libmola_metric_maps.so

4. Use in localization-only mode

In one terminal, launch MOLA-LIO ROS 2 node in localization mode:

ros2 launch mola_lidar_odometry ros2-lidar-odometry.launch.py \
  start_active:=False \
  publish_localization_following_rep105:=False \
  start_mapping_enabled:=False \
  lidar_topic_name:="/ouster/points" \
  imu_topic_name:="/ouster/imu" \
  mola_tf_base_link:="os_sensor" \
  mola_deskew_method:="MotionCompensationMethod::IMU" \
  mola_initial_map_mm_file:=$(pwd)/myMap.mm

Then, either launch again the live LiDAR or ros2 bag play another dataset you want to use for localization.

Finally, two key actions are required: we need to tell MOLA-LIO an initial localization guess and, only after that, enable it so incoming LiDAR scans are actually processed and matched against the prebuilt map.

Note

Other initialization options, such as GNSS-based, are possible via the smoother state estimation.