Kimera-Semantics

Kimera-Semantics

Release News

• Dec 1st 2019: Kimera-Semantics got a complete revamp:
  • Leaner code: no more code dedicated to meshing, we fully re-use Voxblox/OpenChisel instead.
  • New fast method: an order of magnitude faster (took approx 1s before, 0.1s now) than using merged, with minimal accuracy loss for small voxels (it leverages Voxblox' fast approach): you can play with both methods by changing the parameter semantic_tsdf_integrator_type in the launch file. High-res video here.

Publications

We kindly ask to cite our paper if you find this library useful:

• A. Rosinol, M. Abate, Y. Chang, L. Carlone, Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping. IEEE Intl. Conf. on Robotics and Automation (ICRA), 2020. arXiv:1910.02490.

@InProceedings{Rosinol20icra-Kimera,
  title = {Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping},
  author = {Rosinol, Antoni and Abate, Marcus and Chang, Yun and Carlone, Luca},
  year = {2020},
  booktitle = {IEEE Intl. Conf. on Robotics and Automation (ICRA)},
  url = {https://github.com/MIT-SPARK/Kimera},
  pdf = {https://arxiv.org/pdf/1910.02490.pdf}
}

Related publications

Our work is built using Voxblox, an amazing framework to build your own 3D voxelized world:

• Helen Oleynikova, Zachary Taylor, Marius Fehr, Juan Nieto, and Roland Siegwart, Voxblox: Incremental 3D Euclidean Signed Distance Fields for On-Board MAV Planning, in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016.

Which was originally inspired by OpenChisel:

• Matthew Klingensmith, Ivan Dryanovski, Siddhartha Srinivasa, and Jizhong Xiao, Chisel: Real Time Large Scale 3D Reconstruction Onboard a Mobile Device using Spatially Hashed Signed Distance Fields.. Robotics: science and systems (RSS), 2015.

A related work to ours is Voxblox++ which also uses Voxblox for geometric and instance-aware segmentation, differently from our dense scene segmentation, check it out as well!:

• Margarita Grinvald, Fadri Furrer, Tonci Novkovic, Jen Jen Chung, Cesar Cadena, Roland Siegwart, and Juan Nieto, Volumetric Instance-Aware Semantic Mapping and 3D Object Discovery, in IEEE Robotics and Automation Letters, July 2019.

1. Installation

A. Prerequisities

• Install ROS by following our reference, or the official ROS website.

• Install system dependencies:

sudo apt-get install python-wstool python-catkin-tools  protobuf-compiler autoconf
# Change `melodic` below for your own ROS distro
sudo apt-get install ros-melodic-cmake-modules

B. Kimera-Semantics Installation

Using catkin:

…

Finally, compile:

# Compile code
catkin build kimera_semantics_ros

# Refresh workspace
source ~/catkin_ws/devel/setup.bash

2. Usage

First, install Kimera-Semantics, see instructions above.

In Simulation (with semantics)

1. Download the demo rosbag (click here to download) and save it in: ./kimera_semantics_ros/rosbag/kimera_semantics_demo.bag.

2. As a general good practice, open a new terminal and run: roscore

3. In another terminal, launch Kimera-Semantics:

roslaunch kimera_semantics_ros kimera_semantics.launch play_bag:=true

This will launch the rosbag that was downloaded in step 0 and will launch Kimera-Semantics.

1. In another terminal, launch rviz for visualization:

rviz -d $(rospack find kimera_semantics_ros)/rviz/kimera_semantics_gt.rviz

Note: you will need to source your catkin_ws for each new terminal unless you added the following line to your ~/.bashrc file: source ~/catkin_ws/devel/setup.bash # Change bash to the shell you use.

Note 2: you might need to check/uncheck once the Kimera Semantic 3D Mesh left pane topic in rviz to visualize the mesh.

In Euroc dataset (without semantics)

With Kimera-VIO

1. Download a Euroc rosbag: for example V1_01_easy
2. Install Kimera-VIO-ROS.
3. Open a new terminal, run: roscore
4. In another terminal, launch Kimera-VIO-ROS:

roslaunch kimera_vio_ros kimera_vio_ros_euroc.launch run_stereo_dense:=true

The flag run_stereo_dense:=true will do stereo dense reconstruction (using OpenCV's StereoBM algorithm).

1. In another terminal, launch Kimera-Semantics:

roslaunch kimera_semantics_ros kimera_semantics_euroc.launch

1. In yet another terminal, run the Euroc rosbag downloaded in step 0:

rosbag play V1_01_easy.bag --clock

Note 1: Don't forget the --clock flag!

Note 2: Kimera is so fast that you could also increase the rosbag rate by 3 --rate 3 and still see a good performance (results depend on available compute power).

1. Finally, in another terminal, run Rviz for visualization:

rviz -d $(rospack find kimera_semantics_ros)/rviz/kimera_semantics_euroc.rviz

3. FAQ

• Minkindr doesn't compile:

  Catkin ignore the minkindr_python catkin package: touch ~/catkin_ws/src/minkindr/minkindr_python/CATKIN_IGNORE

• How to run Kimera-Semantics without Semantics?

  We are using Voxblox as our 3D reconstruction library, therefore, to run without semantics, simply do:

  bash复制

  roslaunch kimera_semantics_ros kimera_semantics.launch play_bag:=true metric_semantic_reconstruction:=false

• How to enable Dense Depth Stereo estimation

This will run OpenCV's StereoBM algorithm, more info can be found here (also checkout this to choose good parameters):

roslaunch kimera_semantics_ros kimera_semantics.launch run_stereo_dense:=1

This will publish a /points2 topic, which you can visualize in Rviz as a 3D pointcloud. Alternatively, if you want to visualize the depth image, since Rviz does not provide a plugin to visualize a disparity image, we also run a disparity_image_proc nodelet that will publish the depth image to /depth_image.