Hint

Please change branch to Bunker-DVI-Dataset-reg-1 for quick experiment.

Example Dataset:

Download the dataset from Bunker DVI Dataset

benchmark-MAP_LIO-to-HDMapping

Runs the MAP_LIO LiDAR-inertial odometry on a ROS 1 bag file and converts the output to an HDMapping session.

MAP_LIO (E. Vezeteu et al.) is a single-state maximum a posteriori (MAP) LiDAR-inertial estimator with data-driven covariances from local geometric uncertainty. It can additionally fuse a prior airborne (ALS) map and GNSS-INS poses; this benchmark builds and runs the LiDAR-inertial part only (no prior map). MAP_LIO is released for academic research use only, see its repository.

Prerequisites

• Docker
• A ROS 1 bag with a sensor_msgs/PointCloud2 LiDAR topic and a sensor_msgs/Imu topic (ROS 2 bags are automatically converted to ROS 1 format). MAP_LIO reads the Hesai driver's point format: x, y, z, intensity and a per-point timestamp (float64, absolute time in seconds). Clouds with a relative per-point time (float32, seconds after the header stamp), e.g. Livox clouds aggregated by livox_bag_aggregate, are converted to that format automatically.

Step 1 — Clone with submodules

git clone https://github.com/MapsHD/benchmark-MAP_LIO-to-HDMapping.git --recursive
cd benchmark-MAP_LIO-to-HDMapping

Step 2 — Build the Docker image

docker build -t map-lio_noetic .

This installs:

• Ubuntu 20.04 + ROS 1 Noetic
• Eigen3, PCL, GeographicLib and the ROS packages MAP_LIO needs
• Sophus (non-templated version, the commit MAP_LIO pins)
• catkin workspace with map_lio and map_lio_to_hdmapping

Step 3 — Run the pipeline

chmod +x docker_session_run-ros1-map-lio.sh
./docker_session_run-ros1-map-lio.sh /path/to/input.bag /path/to/output/dir

Or with no arguments to use a GUI file selector (requires zenity).

Variable	Default	Meaning
CONFIG_FILE	hesai.yaml	MAP_LIO parameter file: LiDAR/IMU topics, LiDAR-IMU extrinsics, noise, range limits. A file on the host, or the name of one shipped in the image
USE_RVIZ	1	RViz live view

MAP_LIO's sensor setup lives in its parameter file. The image ships two:

Name	Sensor
hesai.yaml	MAP_LIO's own, Hesai LiDAR (/hesai/pandar, /imu/data_raw)
bunker.yaml	Livox Mid-360 with its built-in IMU (/livox/pointcloud, /livox/imu), see overlay/config/bunker.yaml

For your own sensor, copy one of them, set common/lid_topic, common/imu_topic and the LiDAR-IMU extrinsics, and pass it:

CONFIG_FILE=/path/to/my_sensor.yaml \
  ./docker_session_run-ros1-map-lio.sh /path/to/input.bag /path/to/output/dir

Before anything starts, the script checks that both topics of the parameter file are in the bag and looks at the LiDAR point format: Hesai points are used as they are, points with a relative per-point time are converted to the Hesai format first (inside the container, the input bag is not changed), anything else gets a warning (MAP_LIO then does not deskew the scans).

What happens:

The script opens a Docker container with a tmux session containing four panes on window 0 and a control window (window 1, the attach target):

Pane	Role
0	roscore
1	rosbag record /Odometry /cloud_registered
2	roslaunch map_lio map_lio_bench.launch — MAP_LIO reading your bag (+ RViz live view)
3	diagnostics — shows active topics and publishing rates

MAP_LIO does not need rosbag play: it opens the bag itself and processes it as fast as it can. When it reports the end of the bag, the control window stops the recorder, kills all nodes and RViz, and exits tmux. A second Docker run then converts the recorded bag into the HDMapping session format.

Step 4 — Open in HDMapping

Output files appear in <output_dir>/output_hdmapping-MAP_LIO/:

lio_initial_poses.reg
poses.reg
scan_lio_0.laz
...
session.json
trajectory_lio_0.csv
...

Open session.json with the multi_view_tls_registration_step_2 application.

Notes on MAP_LIO

MAP_LIO publishes:

Topic	Type	Meaning
/Odometry	nav_msgs/Odometry	the 6-DoF IMU (body) pose in the world frame, one per scan
/cloud_registered	sensor_msgs/PointCloud2	the registered scan in the world frame, one per scan
/Laser_map	sensor_msgs/PointCloud2	the local map, for display

The converter records /Odometry + /cloud_registered. MAP_LIO publishes the registered scan only while someone subscribes to it, so the recorder is started before MAP_LIO.

Timestamps: the benchmark does not change MAP_LIO's clock (no use_sim_time). MAP_LIO reads the bag directly and stamps every output with the scan's end time taken from the input messages, so the result is in bag time without any translation.

Contact

januszbedkowski@gmail.com