Hello, I have a few questions to ask, thank you:

1. I output GPS data and IMU from the handheld data set_ Correct, it is found that the starting point and the ending point are not at the same point. How to evaluate the accuracy of the method;

2. Slam path is a relative position. How to align it with the ENU of GPS path;

3. The sampling rates of GPS and slam path are different, and the time axis is also misplaced. How does RMSE w.r.t GPS evaluate it.

Hello here, the data is from KITTI dataset, I gave sensor information and internal and external parameters in yaml as required, but every time I open lidar for depth info in params_camera.yaml, the vio part will crash with the following error. Is this a problem with external parameter settings?

After reading your paper and code, thank you very much for your work. But I have the following questions:

1. The laser odometry only uses the result of the visual odometry as the initial optimization value. This should be loosely coupled.
2. Compared with LIO-SAM, LVI-SAM only uses the result of the visual odometry as the initial value of scanning matching, and does not improve the laser odometry. Can the performance of the LVI-SAM really be significantly improved?
3. The results of laser odometry, visual odometry, and IMU pre-integration are used in the paper to perform factor graph optimization. But only the results of the laser odometry and IMU are used in the code. The laser odometry has higher estimation accuracy than the visual odometry, so adding the result of the visual odometry to the factor graph will reduce the optimization accuracy, right?

Hello, when I run LVI Sam on the Jetson NX development board，catkin_make has been completed，but when I tried roslaunch, I encountered the following node death problem

[ INFO] [1636441314.154527795]: ----> Visual Odometry Estimator Started. [ INFO] [1636441314.487589783]: ----> Visual Loop Detection Started. [ INFO] [1636441314.585487422]: ----> Visual Feature Tracker Started. [lvi_sam_visual_odometry-9] process has died [pid 13242, exit code -11, cmd /home/nvidia/catkin_sam/devel/lib/lvi_sam/lvi_sam_visual_odometry __name:=lvi_sam_visual_odometry __log:=/home/nvidia/.ros/log/e982fd3c-412a-11ec-ad02-48b02d3da899/lvi_sam_visual_odometry-9.log]. log file: /home/nvidia/.ros/log/e982fd3c-412a-11ec-ad02-48b02d3da899/lvi_sam_visual_odometry-9*.log [lvi_sam_visual_odometry-9] restarting process process[lvi_sam_visual_odometry-9]: started with pid [13528] [ INFO] [1636441314.871099244]: ----> Visual Odometry Estimator Started. [ INFO] [1636441314.993614341]: ----> Lidar IMU Preintegration Started. [lvi_sam_visual_loop-10] process has died [pid 13249, exit code -11, cmd /home/nvidia/catkin_sam/devel/lib/lvi_sam/lvi_sam_visual_loop __name:=lvi_sam_visual_loop __log:=/home/nvidia/.ros/log/e982fd3c-412a-11ec-ad02-48b02d3da899/lvi_sam_visual_loop-10.log]. log file: /home/nvidia/.ros/log/e982fd3c-412a-11ec-ad02-48b02d3da899/lvi_sam_visual_loop-10*.log [lvi_sam_visual_loop-10] restarting process [ INFO] [1636441315.091523468]: ----> Lidar Cloud Deskew Started. process[lvi_sam_visual_loop-10]: started with pid [13703] [ INFO] [1636441315.169702089]: ----> Lidar Feature Extraction Started. [lvi_sam_visual_feature-8] process has died [pid 13231, exit code -11, cmd /home/nvidia/catkin_sam/devel/lib/lvi_sam/lvi_sam_visual_feature __name:=lvi_sam_visual_feature __log:=/home/nvidia/.ros/log/e982fd3c-412a-11ec-ad02-48b02d3da899/lvi_sam_visual_feature-8.log]. log file: /home/nvidia/.ros/log/e982fd3c-412a-11ec-ad02-48b02d3da899/lvi_sam_visual_feature-8*.log

My environment is: Ubuntu18.04 ROS melodic gtsam4.0.2 ceres1.14.0 opencv4.1.1 pcl1.8

How can I solve the problem？

There will be a significant drop in the z-axis after around 2mins running. After then, LVI-SAM will become totally unstable. Is there anybody who comes across this issue?

Hi,thanks for your great works. the code test well with your data,but it seems the vins program did not work well with my data. I’m confused about what the “question about q_lidar_to_cam q_lidar_to_cam_eigen” mean in the code. hope your reply ,thanks agian. @TixiaoShan

malloc(): memory corruption [lvi_sam_mapOptmization-6] process has died [pid 13598, exit code -6, cmd /home/mwy/lvisam/devel/lib/lvi_sam/lvi_sam_mapOptmization __name:=lvi_sam_mapOptmization __log:=/home/mwy/.ros/log/f460d3d2-9099-11ec-a978-9061ae86e6b5/lvi_sam_mapOptmization-6.log]. log file: /home/mwy/.ros/log/f460d3d2-9099-11ec-a978-9061ae86e6b5/lvi_sam_mapOptmization-6*.log

How to solve this problem? I install the gtsam following your introduction. I am sure I have used cmake -DGTSAM_BUILD_WITH_MARCH_NATIVE=OFF ..
And my pcl version is 1.8.

  void findPosition(double relTime, float *posXCur, float *posYCur, float *posZCur)
  {
      *posXCur = 0; *posYCur = 0; *posZCur = 0;

      // if (cloudInfo.odomAvailable == false || odomDeskewFlag == false)
      //     return;

      // float ratio = relTime / (timeScanNext - timeScanCur);

      // *posXCur = ratio * odomIncreX;
      // *posYCur = ratio * odomIncreY;
      // *posZCur = ratio * odomIncreZ;
  }

As far as I understand, this will result in the pcl being deskewed in rotation only, is this correct?

Thank you for share awesome work.

I can't reproduce result with shared dataset(https://drive.google.com/drive/folders/1q2NZnsgNmezFemoxhHnrDnp1JV_bqrgV?usp=sharing).

It works fine at the beginning. However, at several points, displayed the warning messages and the trajectory was drifted. also, I checked library version. (gtsam-4.0.2, ceres-solver-1.14.0)

Could PC specs have an impact? I installed Linux on Macbook pro A1707. Processo : 2.6 GHz Intel Core i7 (I7-6700HQ) RAM: 16GB GPU: Radeon Pro 450

Warning message: Large bias, reset IMU-preintegration! Large velocity, reset IMU-preintegration!

Hello @TixiaoShan, Thanks for your great work! I couldn't download the test data from the google drive due to it's size is larger than 10G. Could you consider to upload your data to Baidu Netdisk? This may be convenient to people in China.

Hi Tixiao,

I am trying to do some experiments of LVI SAM on the NTU VIRAL public dataset (download page: https://ntu-aris.github.io/ntu_viral_dataset/). Hence, I will include LVI-SAM to the list of applicable methods to NTU VIRAL website.

However LVI SAM diverges quickly after a few minutes. This does not happen if the visual nodes are disabled. Could you please take a look at the configurations and suggest the best configurations?

The forked repository can be found here: https://github.com/brytsknguyen/LVI-SAM

Working examples of LIO SAM and VINs-Mono on NTU VIRAL datasets can be found here: https://github.com/brytsknguyen/LIO-SAM https://github.com/brytsknguyen/VINS-Mono

Hi there!

Does anyone have any thoughts or suggestions on how / whether to use LVI-SAM in more dynamic environments (lots of moving objects including urban environments or indoor scenes)?

I saw previously that LIO-SAM you can utilize Scan Context + Removert and there was a recent paper RF-LIO that focused on object removal, but I would like to utilize the additional VIO data for better map quality while also being able to filter out for moving objects (ideally I would also like to get a track for the moving objects as well or at least the points associated).

Any thoughts on this?

@TixiaoShan
Hi, Thanks for your great work.
I add some functions to save Trajectory in the txt file.
This file is in TUM format, so it could directly visualize using evo tool.

Below is a visualization of the results of the M2DGR dataset using evo tool.

I hope this PR is helpful.

Thanks,

The sentence "The constraints from visual odometry, lidar odometry, IMU preintegration, and loop closure are optimized jointly in the factor graph" is mentioned in related articles. but it seems that visual constrains and inertial constrains are not contained in SAM optimization of mapOptimization.cpp file.

Hi! Thanks for your excellent work. I am running your algorithm, but I do not know how to calibrate fisheye and Lidar. could you please tell me how do you calibrate and which algorithm you use to finish calibration?

Many thanks! :)

@TixiaoShan
Thanks for your great work!
In the LVI-SAM paper, this algorithm adapted LIS(LiDAR inertial system) fail detection.

However, maybe because I'm an newbie, I couldn't find which part of the code it was.
Which part of the code has the LIS fail detection part implemented?

Thanks,