FAST-LIO-SAM-SC-QN

• This repository is a SLAM implementation combining FAST-LIO2 with pose graph optimization and loop closing based on ScanContext, Quatro, and Nano-GICP module
  • ScanContext - a global descriptor for LiDAR point cloud, here it is used as loop candidate pair detection
  • Quatro - fast, accurate and robust global registration which provides great initial guess of transform
  • Quatro module - Quatro as a module, can be easily used in other packages
  • Nano-GICP module - fast ICP combining FastGICP + NanoFLANN
• Note: similar repositories already exist
  • FAST_LIO_LC: FAST-LIO2 + SC-A-LOAM based SLAM
  • FAST_LIO_SLAM: FAST-LIO2 + ScanContext based SLAM
  • FAST_LIO_SAM: FAST-LIO2 + LIO-SAM (not modularized)
  • FAST_LIO_SAM: FAST-LIO2 + LIO-SAM (modularized)
  • FAST_LIO_SAM_QN: FAST-LIO2 + LIO-SAM + Quatro + Nano-GICP (modularized)
• Note2: main code (PGO) is modularized and hence can be combined with any other LIO / LO
  • This repo is to learn GTSAM myself!
  • and as GTSAM tutorial for beginners - GTSAM 튜토리얼 한글 포스팅

Video clip - https://youtu.be/MQ8XxRY472Y

Main difference between FAST_LIO_SAM_QN and FAST_LIO_SAM_SC_QN

• FAST_LIO_SAM_QN sets loop candidate pair as (current keyframe, the closest and old enough keyframe)
• FAST_LIO_SAM_SC_QN gets loop candidate pair from ScanContext

Dependencies

• C++ >= 17, OpenMP >= 4.5, CMake >= 3.10.0, Eigen >= 3.2, Boost >= 1.54
• ROS
• GTSAM >= 4.1.1

  wget -O gtsam.zip https://github.com/borglab/gtsam/archive/refs/tags/4.1.1.zip
  unzip gtsam.zip
  cd gtsam-4.1.1/
  mkdir build && cd build
  cmake -DGTSAM_BUILD_WITH_MARCH_NATIVE=OFF -DGTSAM_USE_SYSTEM_EIGEN=ON ..
  sudo make install -j16
  

• Teaser++

  git clone https://github.com/MIT-SPARK/TEASER-plusplus.git
  cd TEASER-plusplus && mkdir build && cd build
  cmake .. -DENABLE_DIAGNOSTIC_PRINT=OFF
  sudo make install -j16
  sudo ldconfig
  

• tbb (is used for faster Quatro)

  sudo apt install libtbb-dev
  

How to build and use

• Get the code and then build the main code.

  cd ~/your_workspace/src
  git clone https://github.com/engcang/FAST-LIO-SAM-SC-QN --recursive
  
  cd ~/your_workspace
  # nano_gicp, quatro first
  catkin build nano_gicp -DCMAKE_BUILD_TYPE=Release
  
  # Note the option!
  catkin build quatro -DCMAKE_BUILD_TYPE=Release -DQUATRO_TBB=ON
  
  catkin build -DCMAKE_BUILD_TYPE=Release
  . devel/setup.bash
  

• Then run (change config files in third_party/FAST_LIO)

  roslaunch fast_lio_sam_sc_qn run.launch lidar:=ouster
  roslaunch fast_lio_sam_sc_qn run.launch lidar:=velodyne
  roslaunch fast_lio_sam_sc_qn run.launch lidar:=livox
  

Structure

• odom_pcd_cb
  • pub realtime pose in corrected frame
  • keyframe detection -> if keyframe, add to pose graph + save to keyframe queue + generate ScanContext
  • pose graph optimization with iSAM2
• loop_timer_func
  • process a saved keyframe
    • detect loop -> if loop, add to pose graph
• vis_timer_func
  • visualize all (Note: global map is only visualized once uncheck/check the mapped_pcd in rviz to save comp.)