FAST-LIO-SAM

• This repository is a SLAM implementation combining FAST-LIO2 with pose graph optimization and loop closing based on LIO-SAM paper
  • Loop-detection is based on radius search and ICP is used to calc matching
• 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
• 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

KITTI seq 05 top view - (left): FAST-LIO2 (right): FAST-LIO-SAM

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

KITTI seq 05 trajectories - (blue): FAST-LIO2 (green): FAST-LIO-SAM

Note

• For better loop-detection and transform calculation, FAST-LIO-SAM-QN is also coded and opened.
  • It adopts Quatro - fast, accurate and robust global registration which provides great initial guess of transform
  • and Nano-GICP - fast and accurate ICP combining FastGICP + NanoFLANN

Dependencies

• ROS (it comes with Eigen and PCL)
• 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

How to build and use

• Get the code and build

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

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

  roslaunch fast_lio_sam run.launch lidar:=ouster
  roslaunch fast_lio_sam run.launch lidar:=velodyne
  roslaunch fast_lio_sam 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
  • 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.)