MAD-BA

3D LiDAR Bundle Adjustment — from Uncertainty Modelling to Structure Optimization

Accepted RA-L 2025

Build and Run

The MAD-BA was developed and tested on Ubuntu 20.04 with ROS Noetic.

Create Docker Image and Container

• Clone repository

git clone https://github.com/rvp-group/mad-ba.git

• Build docker image and create container:

cd mad_ba && ./build_and_run.sh

Build MAD-BA

• Start and enter container:

docker start mad_ba && docker exec -it mad_ba bash

• Build ROS packages (inside container):

cd /catkin_ws && catkin build -DCMAKE_BUILD_TYPE=Release

Prepare data

• Download example .bag file (inside container):

cd /root/share/dataset/NewerCollege/quad_easy/bag && bash download_quad.sh # quad_easy sequence
cd /root/share/dataset/VBR/Spagna/bag && bash download_spagna.sh # Spagna sequence

• Preprocess .bag file by reordering message using their timestamps and add inital trajectory as nav_msgs/Odometry messages:

cd /root/share/dataset/NewerCollege/quad_easy && python3 process_bag.py  # quad_easy sequence
cd /root/share/dataset/VBR/Spagna && python3 process_bag.py  # Spagna sequence

Run MAD-BA

• Start roscore in background:

screen -dmS roscore roscore

• Run MAD-BA:

roscd mad_ba && rosrun mad_ba main_app -c config/NewerCollege/quad_easy_fast.config # quad_easy sequence
roscd mad_ba && rosrun mad_ba main_app -c config/VBR/spagna.config # Spagna sequence

Output

• Check ./docker_shared/output/ folder:
  • tum - contains the optimized trajectory for each iteration
  • pcd - contains the optimized map for each iteration

Parameters

To modify parameters edit .config file for given sequence, located in /catkin_ws/src/mad_ba/config in Docker container:

"filename":           "path to .bag file"
"topics":             "topics for point cloud and odometry messages"
"clouds_to_process":  "number of point clouds in .bag file"
"decimate_real_data": "decimation - increase for longer sequences"
"iter_num":           "number of BA iterations"