Kimera-PGMO

Kimera-PGMO (Pose Graph and Mesh Optimizer) is an optimizer that takes in a mesh along with robot odometry and loop closure measurements and then optimizes for the trajectory and the mesh simultaneously. This is done by first creating a deformation graph using the pose graph, the mesh, and the time synchronized connections between the pose graph nodes and the mesh vertices. The optimization problem is then optimized using Kimera-RPGO.

If you find this library helpful or use it in your projects, please cite:

@article{rosinol21ijrr-kimera,
	author = {Antoni Rosinol and Andrew Violette and Marcus Abate and Nathan Hughes and Yun Chang and Jingnan Shi and Arjun Gupta and Luca Carlone},
	title ={Kimera: From SLAM to spatial perception with 3D dynamic scene graphs},
	journal = {The International Journal of Robotics Research},
	volume = {40},
	number = {12-14},
	pages = {1510-1546},
	year = {2021},
	doi = {10.1177/02783649211056674}
}

Dependencies

In addition to PCL, OpenCV, and GTSAM, Kimera-PGMO is designed as part of Kimera, so the following Kimera packages and their dependencies are needed:

Kimera-RPGO

pose_graph_tools

Kimera-Semantics

For the immediate dependencies, check out the rosinstall files.

cd ~/catkin_ws/src
wstool merge Kimera-PGMO/install/kimera_pgmo_ssh.rosinstall
wstool update
catkin build kimera_pgmo

Parameters

Mesh Frontend

• horizon is the compressor horizon. We are currently using an geometric octree based mesh simplification technique, vertices that is outside the horizon will no longer be in the octree allowing a new vertex to be places near it.
• output_mesh_resolution resolution of the full mesh. Set to less than the Kimera-Semantics / Voxblox voxel size if you don't want any simplification on the full mesh.
• graph_comppression_method sets the compression method to extract simplified mesh for deformation graph. Recommended: 1.
• full_compression_method sets the compression method for the full mesh. Recommended: 2.
• robot_id can be just set to the default 0 if running single robot. This is really only important for the multirobot case.
• d_graph_resolution resolution of the simplified mesh to be added to the deformation graph.
• log_path path to the folder to save the mesh-frontend log file.
• log_output toggle to log timing and statistics.

Kimera PGMO

• output_prefix path to the folder to save the log file and the optimized mesh and trajectory files.
• robot_id can be just set to the default 0 if running single robot. This is really only important for the distributed multirobot case.
• run_mode toggles the different modes. Set to 0 to receive pose graph and mesh and perform simultaneous pose graph and mesh optimization. Set to 1 to optimize the mesh and subscribe to an optimized trajectory.
• log_output log timing statistics.
• rpgo/ sets various pose graph optimization parameters. See example config.
• add_initial_prior adds a prior factor on first node.
• covariance/ sets the covariance. See example config.

Running Kimera-PGMO

Single robot Kimera

Tesse UHumans2 dataset

In one terminal, launch Kimera-VIO-ROS with stereo dense:

roslaunch kimera_vio_ros kimera_vio_ros_uhumans2.launch

Then launch kimera_pgmo:

roslaunch kimera_pgmo kimera_pgmo.launch dataset:=uHumans2

Launch Kimera-Semantics:

roslaunch kimera_semantics_ros kimera_semantics_uHumans2.launch

For visualization, an rviz configuration is provided:

rviz -d $(rospack find kimera_pgmo)/rviz/uHumans2.rviz

Finally play the rosbag.

rosbag play some_bag.bag --clock --pause

To save the mesh, do

rosservice call /kimera_pgmo/save_mesh

and to save optimized trajectory, do

rosservice call /kimera_pgmo/save_trajectory

the mesh will be saved to ouput_folder/mesh_pgmo.ply and trajectory will be saved to output_folder/traj_pgmo.csv (see launch file)

You can also save the underlying deformation graph to output_folder/pgmo.dgrf

rosservice call /kimera_pgmo/save_dgrf

Loading a mesh and deformation graph

You may also directly load a .ply and .dgrf file to Kimera-PGMO

roslaunch kimera_pgmo kimera_pgmo.launch dataset:=uHumans2

Start rviz:

rviz -d $(rospack find kimera_pgmo)/rviz/uHumans2.rviz

And load mesh and deformation graph:

rosservice call /kimera_pgmo/load_graph_mesh '{robot_id: 0, dgrf_file: /home/yunchang/catkin_ws/src/kimera_pgmo/kimera_pgmo/log/pgmo.dgrf, ply_file: /home/yunchang/catkin_ws/src/kimera_pgmo/kimera_pgmo/log/mesh_pgmo.ply}'

Developer notes

Running the Unit-tests:

roscd kimera_pgmo
catkin run_tests --no-deps --this
catkin_test_results ~/catkin_ws/build/kimera_pgmo/

You can also run individual tests for example:

rostest kimera_pgmo test_mesh_frontend.test --text

Or

rosrun kimera_pgmo kimera_pgmo-test_deformation_graph

Misc Note

One thing to note if a developer is working with GTSAM and want to add other factors into the system is that here we specify different prefixes for different types of nodes in the deformation graph, take a look at utils/CommonFunctions.h for reference. By prefix we mean the key character as described here.