snt-arg / lidar_situational_graphs

Repository to generate in real-time Situational Graphs (S-Graphs) for robot pose and map optimization using 3D LiDAR Data

LiDAR S-Graphs

LiDAR Situational Graphs (S-Graphs) is a ROS2 package for generating in real-time four-layered hierarchical factor graphs representing a scene graph using 3D LiDAR which includes Keyframes registring the robot poses, Walls which map wall planes, Rooms Layer constraining the wall planes using 4 wall-room or 2 wall-room factors, Floors constraining the rooms within a given floor level. It also supports several graph constraints, such as GPS, IMU acceleration (gravity vector), IMU orientation (magnetic sensor). We have tested this package mostly with Velodyne (VLP16) sensors in structured indoor environments. This work is a fork of hdl_graph_slam which as previously in ROS1.

📜 Table of contents

• 📖 Published Papers
• ⚙️ Installation
  • 📦 Installation From Source
    • ROS1 Optional
  • 🐳 Docker
• 🚀 Examples on Datasets
• 🛠️ Run S_Graphs On Your Data
• 🤖 ROS Related
  • 📥 Subscribed Topics
  • 📤 Published Topics
  • 🔄 ROS Services
  • ⚙️ ROS Parameters
  • 🌐 Published TFs
• 🧪 Unit Tests

📖 Published Papers

S-Graphs+: Real-time Localization and Mapping leveraging Hierarchical Representations

@ARTICLE{10168233,
 author={Bavle, Hriday and Sanchez-Lopez, Jose Luis and Shaheer, Muhammad and Civera, Javier and Voos, Holger},
 journal={IEEE Robotics and Automation Letters},
 title={S-Graphs+: Real-Time Localization and Mapping Leveraging Hierarchical Representations},
 year={2023},
 volume={8},
 number={8},
 pages={4927-4934},
 doi={10.1109/LRA.2023.3290512}}

Situational Graphs for Robot Navigation in Structured Indoor Environments

  @ARTICLE{9826367,
    author={Bavle, Hriday and Sanchez-Lopez, Jose Luis and Shaheer, Muhammad and Civera, Javier and Voos, Holger},
    journal={IEEE Robotics and Automation Letters},
    title={Situational Graphs for Robot Navigation in Structured Indoor Environments},
    year={2022},
    volume={7},
    number={4},
    pages={9107-9114},
    doi={10.1109/LRA.2022.3189785}}

⚙️ Installation

Note

S-Graphs+ was only tested on Ubuntu 20.04, ROS2 Foxy, Humble Distros. We strongly recommend using cyclone_dds instead of the default fastdds.

📦 Installation From Source

Important

Before proceeding, make sure you have rosdep installed. You can install it using sudo apt-get install python3-rosdep In addition, ssh keys are needed to be configured on you GitHub account. If you haven't yet configured ssh keys, follow this tutorial

1. Update Rosdep:

rosdep init && rosdep update --include-eol-distros

2. Create a ROS2 workspace for S-Graphs

mkdir -p $HOME/workspaces && cd $HOME/workspaces

3. Clone the S-Graphs repository into the created workspace

git clone [e-mail]:snt-arg/lidar_situational_graphs.git -b feature/ros2 s_graphs 

Important

If you have Nvidia GPU please install CUDA from this link. This code has only been tested with CUDA 11.8. If you dont have CUDA S-Graphs will use CPU only.

4. Install required dependencies. Change $ROS_DISTRO to your ros2 version.

cd s_graphs && source /opt/ros/$ROS_DISTRO/setup.sh && ./setup.sh

Note

If you want to compile with debug traces (from backward_cpp) run:

colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=RelWithDebInfo

Optional ROS1 Install (Old Version of Room Segmentation)

[!NOTE] This is an optional older version of room segmentation algorithm which requires ROS1 noetic. There is no hard dependecy on this package so you can easily ignore this step.

Download ROS Bridge

source /opt/ros/foxy/setup.bash && sudo apt install ros-foxy-ros1-bridge

Installation on ROS1

[!IMPORTANT] Before following the instructions from below, ensure that you are in a fresh terminal, without ROS2 sourced.

1. Create a ROS1 workspace for S-Graphs

mkdir -p $HOME/workspaces/s_graphs_ros1_ws/src && cd $HOME/workspaces/s_graphs_ros1_ws/src && source /opt/ros/noetic/setup.bash

2. Clone the S-Graphs repository into the created workspace

git clone [e-mail]:snt-arg/lidar_situational_graphs.git -b feature/ros2 s_graphs

3. Install required dependencies using vcstool

cd s_graphs && vcs import --recursive ../ < .rosinstall_ros1

4. Install required ROS packages

cd ../../ && rosdep install --from-paths src --ignore-src -y -r

5. Install pcl_ros

sudo apt install ros-noetic-pcl-ros

6. Build workspace

[!IMPORTANT] Make sure s_graphs_ros1_ws is built in Release otherwise the room extraction won't work properly.

catkin config --cmake-args -DCMAKE_BUILD_TYPE=Release && catkin build

🐳 Docker

Build Docker Image

1. Create a ROS2 workspace for S-Graphs

mkdir -p $HOME/workspaces && cd $HOME/workspaces

2. Change directory to where Dockerfile is located in s_graphs

git clone [e-mail]:snt-arg/lidar_situational_graphs.git -b feature/ros2 s_graphs && cd $HOME/workspaces/s_graphs/docker/foxy_noetic

3. Build image

docker build -t sntarg/s_graphs .

Note

There are two docker files, one for foxy and another for humble. The above commands build the foxy image, you use the same commands to build the humble image if needed.

🚀 Example on Datasets

Warning

For execution of the experiments we use mprocs, which makes the process of launching different processes easier.

Real Dataset

Important

Download real dataset using this link and store it in the folder ~/Downloads/real, the below mprocs script will not work otherwise.

cd $HOME/workspaces/s_graphs && mprocs --config .real_mprocs.yaml

Virtual Dataset

Important

Download virtual dataset using this link and store it in the folder ~/Downloads/virtual, the below mprocs script will not work otherwise.

cd $HOME/workspaces/s_graphs && mprocs --config .virtual_mprocs.yaml

Running S_Graphs with Docker

Note

This tutorial assumes that you have followed the instructions to setup docker in section 🐳 Docker

1. Create a container for the s_graphs image.

docker run -dit --volume=/tmp/.X11-unix:/tmp/.X11-unix:rw --network=host -e DISPLAY=$DISPLAY --name s_graphs_container sntarg/s_graphs

2. Download the dataset you desire from above to your local machine.

3. Move the rosbag inside docker container

docker cp ~/Downloads/real s_graphs_container:/root/Downloads/real # For real dataset
# OR
docker cp ~/Downloads/virtual s_graphs_container:/root/Downloads/virtual # For virtual dataset

4. Execute the container

docker exec -ti s_graphs_container bash

Important

If rviz2 doesnt open inside the docker, do xhost + in a terminal of your pc and then relaunch the mprocs command inside docker.

3. Run mprocs

mprocs_real # To run on a real robot or a real dataset
# OR
mprocs_virtual # To run on a simulation or virtual dataset

Note

Press reset on rviz2 once in a while when running S-Graphs to avoid freezing effect caused by rviz2 on foxy.

🛠️ Run S_Graphs On Your Data

1. Define the transformation between your sensors (LIDAR, IMU, GPS) and base_link of your system using static_transform_publisher (see line, s_graphs_launch.py). All the sensor data will be transformed into the common base_link frame, and then fed to the SLAM algorithm. Note: base_link frame in virtual dataset is set to base_footprint and in real dataset is set to body. You can set the frames, topics for your dataset easily during the launch execution as follows:

    ros2 launch lidar_situational_graphs s_graphs_launch.py compute_odom:=true lidar_topic:=/rs_lidar/points

3. If you have an odometry source convert it to base ENU frame, then set the arg compute_odom to false in s_graphs_ros2_launch.py and then remap odom topic in s_graphs_node like

  ros2 launch lidar_situational_graphs s_graphs_launch.py compute_odom:=false lidar_topic:=/rs_lidar/points odom_topic:=/odom

Note

If you want to visualize the tfs correctly from your odom source, you MUST provide a tf from the odom to base_link frame.

🤖 ROS Related

📥 Subscribed Topics

s_graphs node

Topic name	Message Type	Description
/odom	nav_msgs/Odometry	The odometry from the robot.
/filtered_points	sensor_msgs/PointCloud2	The filtered data from the LiDAR sensor.

room_segmentation node

Topic name	Message Type	Description
/voxblox_skeletonizer/sparse_graph	visualization_msgs/MarkerArray	Represents the free space where the robot can go to. Also known as free-space clusters.
/s_graphs/map_planes	s_graphs/PlanesData	Planes seen by the current robot keyframe.

floor_plan node

Topic name	Message Type	Description
/s_graphs/all_map_planes	visualization_msgs/MarkerArray	All the planes that have been seen by the robot.

📤 Published Topics

s_graphs node

Topic name	Message Type	Description
/s_graphs/markers	visualization_msgs/MarkerArray	These markers represent the different s_graphs layers.
/s_graphs/odom2map	geometry_msgs/TransformStamped	The estimated drift of the robot within its map frame (world).
/s_graphs/odom_pose_corrected	geometry_msgs/PoseStamped	The optimized/drift-free pose of the robot once odom2map is applied.
/s_graphs/odom_path_corrected	nav_msgs/Path	The optimized/drift-free pose path of the robot once the odom2map is applied.
/s_graphs/map_points	sensor_msgs/PointCloud2	The projected 3D points using the optimized robot pose.
/s_graphs/map_planes	s_graphs/PlanesData	Planes seen by the current robot keyframe.
/s_graphs/all_map_planes	s_graphs/PlanesData	All the planes that have been seen by the robot.

room_segmentation node

Topic name	Message Type	Description
/room_segmentation/room_data	s_graphs/RoomsData	Contains all the necessary information about the rooms on a given floor.

floor_plan node

Topic name	Message Type	Description
/floor_plan/floor_data	s_graphs/RoomData	Contains all the necessary information about each floor.

🔄 ROS Services

Topic name	Message Type	Description
/s_graphs/dump	s_graphs/DumpGraph	Save all the internal data (point clouds, floor coeffs, odoms, and pose graph) to a directory.

Topic name	Message Type	Description
/s_graphs/save_map	s_graphs/SaveMap	Save the generated 3D map as a PCD file.

⚙️ ROS Parameters

All the configurable parameters are listed in config folder as ros params.

🌐 Published TFs

• map2odom: The transform published between the map frame and the odom frame after the corrections have been applied.

• The entire tf_tree for the virtual experiment can be seen in the figure below.

🧪 Unit Tests

Some unit tests are available. In case you want to add additional tests, run the following command:

colcon test --packages-select s_graphs --event-handler=console_direct+