RS-SLAM

Authors: T. Ran, L. Yuan, D. Tang
This is a semantic SLAM system that is robust in dyanmic environments.

1. This project is built on ORB-SLAM2, so the Thirdparty and Vocabulary in ORB-SLAM2 should be copyed into rs-slam/. Then compiling the DBoW2 and g2o and uncompressing the ORBvoc.
2. Putting the whole project into the ROS workspace and running catkin_make to compile it. The RGBD node as well as a semantic_cloud node will be generated.
3. Download the segmentation model in model trained on sunrgbd and put them in semantic_slam/models/.
4. Runing the two ROS node to subscibe the image tpoic.
5. Running a .bag file in TUM3 database to publish rgb and depth images or the openni driver if you have a RGB-D camera.

Dependencies

1. Pytorch 0.4.0 is required for semantic segmentation.
2. Octomap is required for map construction.

Acknowledgement

This work cannot be done without many open source projets. Special thanks to
semantic_slam
ORB_SLAM2
ORB_SLAM2_SSD_Semantic

License

This project is released under a GPLv3 license.

Contact us

For any issues, please feel free to contact Teng Ran: rantengsky@163.com