vsgp_pcl

Light-Weight Pointcloud Representation with Variational Sparse Gaussian Process

Building

• Clone the repository as your workspace

git clone https://github.com/mahmoud-a-ali/vsgp_pcl.git

• Compile it using catkin build

cd vsgp_pcl
catkin build
source devel/setup.bash

Usage

1. Launch the simulated cpr_inspection environment which contains lake, pipe, grass, and tunnel

roslaunch jackal_gazebo cpr_inspection.launch 

2. Run occupancy surface which convert pointcloud to an occupancy surface

roslaunch vsgp_gen_mdl occ_surface.launch 

3. Check that only the base node uses GPU where the scout node uses CPU

# check that the following line is commented for `base` node and uncommented for `scout` node
os.environ["CUDA_VISIBLE_DEVICES"]="-1" # disable GPU

4. Run the scout node which trains an VSGP and publish only the inducing points to the base node

roslaunch vsgp_gen_mdl scout.launch

5. Run the base node which construct back the pointcloud using the inducing points send by the scout node

roslaunch vsgp_gen_mdl base.launch

If using two PCs, one as the Scout robot and the other as the Base station

1. Configure the two PCs to communicate to each other using ROS. you can make either Scout or Base as the ROS_MASTER

• On the Scout robot, update .bashrc file by adding

export ROS_MASTER_URI=http://Scout_IP:11311
export ROS_HOSTNAME=Scout_IP

• On the Base station, update .bashrc file by adding

export ROS_MASTER_URI=http://Scout_IP:11311
export ROS_HOSTNAME=Base_IP

• Note: this configuration making the scout as the ROS_MASTER

2. On the Scout PC run the following:

roslaunch jackal_gazebo cpr_inspection.launch
roslaunch vsgp_gen_mdl occ_surface.launch
roslaunch vsgp_gen_mdl scout.launch

3. On the Base PC run the following:

roslaunch vsgp_gen_mdl base.launch

If using real robot as the Scout

• Replace the command to launch the simulator with the launch file that runs the robot hardware including LiDAR driver and localization node
• Update pcl_topic and localization_topic in the occ-surface.launch file

Paper: Light-Weight Pointcloud Representation with Variational Sparse Gaussian Process

Citation

@inproceedings{ali2023light,
  title={Light-weight pointcloud representation with sparse gaussian process},
  author={Ali, Mahmoud and Liu, Lantao},
  booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={4931--4937},
  year={2023},
  organization={IEEE}
}

video