yichencityu / sc-lego-loam

LiDAR SLAM: Scan Context + Lego-LOAM

SC-LeGO-LOAM

• LiDAR SLAM: Scan Context (18 IROS) + Lego-LOAM (18 IROS)
• This repository is an example use-case of Scan Context C++ , the LiDAR place recognition method, for LiDAR SLAM applications.
• Just include Scancontext.h. For details see the file mapOptmization.cpp.

Features

• Light-weight: a single header and cpp file named "Scancontext.h" and "Scancontext.cpp"
  • Our module has KDtree and we used nanoflann . nanoflann is an also single-header-program and that file is in our directory.
• Easy to use: A user just remembers and uses only two API functions; makeAndSaveScancontextAndKeys and detectLoopClosureID.
• Fast: The loop detector runs at 10-15Hz (for 20 x 60 size, 10 candidates)

Examples

• Video 1: DCC (MulRan dataset)
• Video 2: Riverside (MulRan dataset)

Scan Context integration

• For implementation details, see the mapOptmization.cpp; all other files are same as the original LeGO-LOAM.
• Some detail comments
  • We use non-conservative threshold for Scan Context's nearest distance, so expect to maximise true-positive loop factors, while the number of false-positive increases.
  • To prevent the wrong map correction, we used Cauchy (but DCS can be used) kernel for loop factor. See mapOptmization.cpp for details. (the original LeGO-LOAM used non-robust kernel). We found that Cauchy is emprically enough.
  • We use both two-type of loop factor additions (i.e., radius search (RS)-based as already implemented in the original LeGO-LOAM and Scan context (SC)-based global revisit detection). See mapOptmization.cpp for details. SC is good for correcting large drifts and RS is good for fine-stitching.
  • Originally, Scan Context supports reverse-loop closure (i.e., revisit a place in a reversed direction) and examples in here (py-icp slam) . Our Scancontext.cpp module contains this feature. However, we did not use this for closing a loop in this repository because we found PCL's ICP with non-eye initial is brittle.

How to use

• Place the directory SC-LeGO-LOAM under user catkin work space
• For example,

  cd ~/catkin_ws/src
  git clone https://github.com/irapkaist/SC-LeGO-LOAM.git
  cd ..
  catkin_make
  source devel/setup.bash
  roslaunch lego_loam run.launch
  

MulRan dataset

• If you want to reproduce the results as the above video, you can download the MulRan dataset and use the ROS topic publishing tool .

Dependencies

• All dependencies are same as LeGO-LOAM (i.e., ROS, PCL, and GTSAM).