VINS has been completely reconstructed and rewritten using C object-oriented, and supports stereo or stereo IMU.

• build

$\quad$ The test completely passed on Ubuntu20.04 system

mkdir -p catkin_ws/src
cd catkin_ws/src
git clone https://github.com/weihaoysgs/vins-fast.git
cd ..
catkin_make -j

• run

source devel/setup.zsh # or source devel/setup.bash
roslaunch vins rviz.launch
# open new terminal
./build/vins-fast/vins/test_feature_track \
--config_file_path=/home/xx/catkin_ws/src/vins-fast/vins/config/euroc/euroc_stero.yaml

• parameter

$\quad$ maybe you need to change some parameter value in your yaml config file.

output_path: "/home/xx/catkin_ws/" # the vio trajectory result will save as TUM format
output_file_name: "vio.txt"

ros_bag_path: "/home/xx/dataset/euroc/MH_05_difficult.bag" 

cam0_calib: "/home/xx/catkin_ws/src/vins-fast/vins/config/euroc/cam0_mei.yaml"
cam1_calib: "/home/xx/catkin_ws/src/vins-fast/vins/config/euroc/cam1_mei.yaml"

The results of running on the Euroc dataset MH_05_difficult in stereo only mode. The accuracy is evaluated by the EVO tool, the RMSE is 0.40

Generate simulate IMU data through vio-data-simulation, verify whether the pre-integration results in the program are consistent with the normal integration results, and verify the correctness of the Jacobian matrix. You can even generate an entire VIO simulation dataset to verify your slam algorithm.

The stereo IMU mode is test success, the evo_ape result as follows, the RMSE is 0.240860.

   max	0.400143
  mean	0.228765
median	0.221756
   min	0.063146
  rmse	0.240860
   sse	64.453061
   std	0.075366

• VINS-Fusion
• Vio-data-simulation

• Add IMU to opt
• Add deep method
• Add other sensor