FAST-LIO-SAM-QN

• This repository is a SLAM implementation combining FAST-LIO2 with pose graph optimization and loop closing based on Quatro and Nano-GICP module
  • 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)
• 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

    
    

Computational complexity
in KITTI seq. 05 with i9-10900k CPU

• FAST-LIO-SAM: max 118% CPU usage, 125 times of ICP, 124.9ms consumption on average
• FAST-LIO-SAM-N (only Nano-GICP): max 164% CPU usage, 130 times of ICP, 61.9ms consumption on average
• FAST-LIO-SAM-QN:
  • Advanced matching - max 325% CPU usage, 85 times of ICP, 140ms consumption on average
  • Optimized matching (with max 200 correspondences downsampling) - max 569% CPU usage, 90 times of ICP, 128.6ms consumption on average
• Note: loop_timer_func runs at fixed basic/loop_update_hz. So how many times of ICP occurred can be different depending on the speed of matching methods.

KITTI seq 05 top view - (left): FAST-LIO2 (middle): FAST-LIO-SAM (bottom): FAST-LIO-SAM-QN

KITTI seq 05 side view - (top): FAST-LIO2 (middle): FAST-LIO-SAM (bottom): FAST-LIO-SAM-QN

Dependencies

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

  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

• Get the code and then build the main code.

  cd ~/your_workspace/src
  git clone https://github.com/engcang/FAST-LIO-SAM-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 -DQUATRO_DEBUG=OFF
  catkin build -DCMAKE_BUILD_TYPE=Release
  . devel/setup.bash

How to run

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

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

• In particular, we provide a preset launch option for specific datasets:

  roslaunch fast_lio_sam_qn run.launch lidar:=kitti
  roslaunch fast_lio_sam_qn run.launch lidar:=mulran
  roslaunch fast_lio_sam_qn run.launch lidar:=newer-college20

Structure

• odom_pcd_cb
  • pub realtime pose in corrected frame
  • keyframe detection -> if keyframe, add to pose graph + save to keyframe queue
  • 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.)

Memo

• Quatro module fixed for empty matches
• Quatro module is updated with optimizedMatching which limits the number of correspondences and increased the speed