Hi @engcang! Nice integration, your videos look pretty good. I have been trying to replay what you did with LIO-SAM repo but using another robot. I am using a differential drive robot model (https://github.com/sanuann/DifferentialDriveRobot) where I attached a Velodyne VLP-16 (https://github.com/lmark1/velodyne_simulator) and a IMU (libgazebo_ros_imu.so).

I can't get a good reconstruction. Have you seen this behavior? I have found that it could be related to the extrinsic parameters in params.yaml but I believe I'm setting the values correctly.

I attached the Velodyne into the robot chassis using the file myrobot.xacro from https://github.com/sanuann/DifferentialDriveRobot

<xacro:include filename="$(find velodyne_description)/urdf/VLP-16.urdf.xacro"/>
  <VLP-16 parent="chassis" name="velodyne" topic="/velodyne_points" hz="10" samples="440">
    <origin xyz="0 0 0.0" rpy="0 0 0" />
  </VLP-16>

I attached the IMU into the velodyne body frame and defined an update rate of 500Hz in the file myrobot.xacro from https://github.com/sanuann/DifferentialDriveRobot

<gazebo>
    <plugin name="imu_plugin" filename="libgazebo_ros_imu.so">
      <alwaysOn>true</alwaysOn>
      <bodyName>velodyne</bodyName>
      <topicName>imu</topicName>
      <serviceName>imu_service</serviceName>
      <gaussianNoise>0.0</gaussianNoise>
      <updateRate>500.0</updateRate>
    </plugin>
  </gazebo>

Thus, all sensors are in the same body frame, velodyne. That is why I used this Extrinsics

# Extrinsics (lidar -> IMU)
  extrinsicTrans:  [0.000, 0.000, 0.0]
  extrinsicRot: [1, 0, 0,
                  0, 1, 0,
                  0, 0, 1]
  extrinsicRPY: [1,  0, 0,
                 0, 1, 0,
                 0, 0, 1]

but can't get a good reconstruction. I also tested moving the VLP-16 0.4m above the chassis and modified the Extrinsics to [0.000, 0.000, -0.438] (gotten from rosrun tf tf_echo velodyne chassis) but I have the same behavior.

Any suggestion will be appreciated. Thanks!

Great video! I had try R3Live with similar or worse results, mainly due non structural environments (a.k.a agricultural environments).
Another option that I see that you have no try in your github is FAST_LIO_LC who implement loop closure. and GPS support (non mandatory) I'm thinking to implement it in R3Live but I'm not sure why do the team cast all the data to different point clouds (if you look at your topics there are 100 of color point clouds). Do you know why?

Hi, I watched your video, it's very impressive! But I'm not familiar with ROS, I don't know how to create the simulated data, is there any blog or teach documentation?

By the way, I guess your vio is not running well on lvi-sam, since I see the vio point cloud is always behind the robot, but from the image, I think the vio point cloud should be in front of the robot. This maybe ascribed to the configuration in params_camera.yaml, lidar_to_cam_* parameters. I also run into this problem like many other researchers https://github.com/TixiaoShan/LVI-SAM/issues/19 and https://github.com/TixiaoShan/LVI-SAM/issues/21