PL-inertial-slam

Reference works

1. PL-SLAM https://github.com/rubengooj/pl-slam.git
2. VIORB https://github.com/jingpang/LearnVIORB.git
3. VINS-MONO https://github.com/HKUST-Aerial-Robotics/VINS-Mono

Thankyou for their outstanding work. My code is basically copied from them. The basic framework is from PL-SLAM, the IMU preintegration is from VIORB, the marginalization is from VINS-MONO

Required Library

1. g2o
2. OpenCV 3.X.X
3. Eigen3
4. Boost
5. yaml
6. mrpt

The library I use is basically the same as PL-slam. Please refer to the main page of PL-slam for more information. But I put StVO-PL directly into the program, so this library does not need to be installed.

How To Run

I only test on the Euroc dataset of V1_easy, so I take this dataset as example.

0. Set the variable DAFAULT_USE_MARG in the cmakelists.txt to choose if use marginalization in the localmapping thread

1. Use scripts ./build.sh to compile programs

2. Put the dataset_params.yaml under the config/dataset_params directory into the dataset V1_easy root directory

3. Unzip voc.tar.gz under the vocabulary directory

4. Change value of vocabulary_p and vocabulary_l in the file of config_euroc.yaml under the config/config directory

5. Set dataset environment variables and run program. For example, my dataset V1_easy is in the directory /home/xc/Euroc/V1_easy, Run the program with the following command:

   1. export DATASETS_DIR=/home/xc
   2. cd build
   3. ./plslam_dataset Euroc/V1_easy -c ../config/config/config_euroc.yaml

Result

Compare with the origin PL-SLAM

All results are compared with ground truth using the tool of EVO.

The left picture is the result of pl-slam with IMU and marginalization.

The mid picture is the result of pl-slam with IMU but no marginalization.

The right picture is the result of original pl-slam.

Remarks

1. The program may have bugs like (segment fault), if you happen to meet it, just run program again.

2. This is just the code that I used to learn vio. If there is any mistake, please forgive me