1. slam2d

• scan match
• correspondence
• scan-to-map match
• map update

1.1 scan-to-scan match

ICP/PLICP method is used for scan-to-scan match, and ceres is used as the nonlinear solver with point-to-line contraints.

1.2 correspondence

KNN method of pcl lib is used to find point correspondence

1.3 scan-to-map match

find local optimal with random search

1.4 map update

OpenCV function line and circle is used to update grid map.

2. Building and Running

2.1 Building

cd catkin_ws/src
git clone [email protected]:libing64/slam2d.git
cd ..
catkin_make -DCATKIN_WHITELIST_PACKAGES="slam2d"

2.2 Running

soure devel/setup.bash
roslaunch slam2d slam2d.launch

play your rosbag

rosbag play youbag

cartographer dataset:

https://google-cartographer-ros.readthedocs.io/en/latest/data.html

2.3 Enviroment

• Ubuntu20.04
• ros noetic
• OpenCV
• pcl
• Eigen
• ceres

3. TODO

• [ ] how to reduce the drift
• [ ] how to improve the mapping accurcy
• [x] scan-to-map match
• [ ] keyframe selection to reduce drift
• [ ] dynamic range 2d map
• [x] search local optimal
• [ ] record gif with byzana-record
• [ ] sparse pose adjust
• [ ] loop closure

4. Comparison of 2D SLAM

4.1 Hector SLAM

• scan-to-scan match and scan-to-map match, Gaussian-Newton
• need high rate scan data
• 2.5D SLAM
• multi-resolution map to avoid local minima

4.2 GMapping

• Rao-Blackwellized Particle filter approach

4.3 Karto SLAM

• graph-based approach
• landmark based
• Cholesky decomposition for solving sparse linar system
• loop closure
• sparse pose adjustment(just like bundle adjustment)

4.4 CoreSLAM

• ros wrapper of tinySLAM(200 lines-of-code)
• partical filter based approach

4.5 LagoSLAM

• Graph optimization -> Linear approximation of Graph optimization
• linear so no initial value needed

5. map comparision

• grid_map https://github.com/ANYbotics/grid_map
• costmap_2d http://wiki.ros.org/costmap_2d
• OctoMap https://octomap.github.io/

6. Reference

• An ICP variant using a point-to-line metric
• A flexible and Scalable SLAM System with Full 3D motion Estimation
• Real-Time Loop Closure in 2D LIDAR SLAM
• Grid-based Scan-to-Map Matching for Accurate 2D Map Building