Public place to ask questions, post and discuss issues

[laboshinl]

[laboshinl]

The mapping is working, but in some point the pointcloud suddenly returns to the initial position and I get this error:

  laserOdometry: /build/buildd/pcl-1.7-1.7.1/kdtree/include/pcl/kdtree/impl/kdtree_flann.hpp:136: int pcl::KdTreeFLANN<PointT, Dist>::nearestKSearch(const PointT&, int, std::vector<int>&, std::vector<float>&) const [with PointT = pcl::PointXYZI; Dist = flann::L2_Simple<float>]: Assertion `point_representation_->isValid (point) && "Invalid (NaN, Inf) point coordinates given to nearestKSearch!"' failed.
    [laserOdometry-2] process has died [pid 10107, exit code -6, cmd /home/vicky/catkin_ws/devel/lib/loam_velodyne/laserOdometry __name:=laserOdometry __log:=/home/vicky/.ros/log/580f7db4-ffc0-11e5-aee2-74d4358625bb/laserOdometry-2.log].
    log file: /home/vicky/.ros/log/580f7db4-ffc0-11e5-aee2-74d4358625bb/laserOdometry-2*.log
    [laserOdometry-2] restarting process

I think when a keypoint from the registration process is not found, then the node laserOdometry node crashed and restart, then the mapping does not work because it does not continue in the same position as before, however it start again from the initial position.

When you run your example:

roslaunch ~/catkin_ws/src/loam_velodyne/launch/loam_velodyne.launch
rosbag play ~/Downloads/nsh_indoor_outdoor.bag 

do you obtain the same error?

How can I avoid this error? Many thanks in advance.

I have the same errors with the nsh bag. Works great with sample VLP16 data

[laboshinl]

Is this a version you wrote or is it the original implementation or did you piece it back together from the help pages? Did you write the package and launch file?

I've created package and launch file. I've also modified scanRegistration.cpp to get it to work with my VLP16 (I think originally it was for HDL32 )

[ZepherusFF]

@laboshinl Do you know if it saves the map generated anywhere? and if not, are you planning on modifying it so it does?

[DamonMIN]

When I build with catkin "source ~catkin_ws/devel/setup.bash", there is an error "bash: ~catkin_ws/devel/setup.bash: No such file or directory". Please help me to solve it. Thank you. @laboshinl

[ZepherusFF]

Hi @DamonMIN did you setup your catkin workspace correctly based on the ROS documentation ? Do you have the file in that directory ?

[DamonMIN]

Hi@ZepherusFF I have the "setup.bash" file in the folder ~catkin_ws/devel. The following text is the content of this file.

////////////////////////////////////////////////////////////////////
#!/usr/bin/env bash
# generated from catkin/cmake/templates/setup.bash.in

CATKIN_SHELL=bash

# source setup.sh from same directory as this file
_CATKIN_SETUP_DIR=$(builtin cd "`dirname "${BASH_SOURCE[0]}"`" > /dev/null && pwd)
. "$_CATKIN_SETUP_DIR/setup.sh"
//////////////////////////////////////////////////////////////////

[laboshinl]

@DamonMIN You have missed "/"

laboshinl@rtc ~/catkin_ws $ source ~catkin_ws/devel/setup.bash 
bash: ~catkin_ws/devel/setup.bash: No such file or directory

laboshinl@rtc ~/catkin_ws $ source ~/catkin_ws/devel/setup.bash 

[DamonMIN]

@laboshinl

In second terminal play sample velodyne data from VLP16 rosbag:

rosbag play ~/Downloads/2014-11-10-10-48-31_Velodyne-VLP_20Hz_Hecker Pass.pcap
[ INFO] [1461760124.789037514]: Opening /home/aicrobo/Downloads/2014-11-10-10-48-31_Velodyne-VLP_20Hz_Hecker
[FATAL] [1461760124.789397586]: Error opening file: /home/aicrobo/Downloads/2014-11-10-10-48-31_Velodyne-VLP_20Hz_Hecker

read from velodyne VLP16 sample pcap:

roslaunch velodyne_pointcloud VLP16_points.launch pcap:="/home/aicrobo/Downloads/2014-11-10-10-48-31_Velodyne-VLP_20Hz_Hecker Pass.pcap"
[VLP16_points.launch] is neither a launch file in package [velodyne_pointcloud] nor is [velodyne_pointcloud] a launch file name
The traceback for the exception was written to the log file

It seems that the launch file can't find or recognize the velodyne data (The data was downloaded from VLP16 sample pcap).

[laboshinl]

You should compile velodyne driver https://github.com/ros-drivers/velodyne/tree/master/velodyne_pointcloud

[ZepherusFF]

@DamonMIN For the VLP16 you need to compile yourself the driver

[FernandoArrigorriaga]

Is there any way to export the data or is just for visualization?. By the way, works great.

[laboshinl]

@Fazt You can save registered clouds to pcd and open with ccViewer rosrun pcl_ros pointcloud_to_pcd input:=/velodyne_cloud_registered

[drscopus]

It seems that I can't run it successfully. I used the "VLP16 sample pcap" data sets. Should I calibrate the Velodyne sensor at first? At first, the 3D data points couldn't be displayed in rviz, only two odom topics /laser_odom_to_init and /integrated_to_init could be displayed. The locations of these odom topics are strange. Initially, they start at a same position. They are getting further in the distance. And the orientations of these two topics are also very strange. Then I found if the displayed style of pointcloud2 is adjusted from "points" to "squares", the 3D data points could be displayed. However, I couldn't see the constructed 3D map in rviz. Looking forwards to any great advices. Best!

[laboshinl]

@drscopus

only two odom topics /laser_odom_to_init and /integrated_to_init could be displayed. The locations of these odom topics are strange. Initially, they start at a same position. They are getting further in the distance. And the orientations of these two topics are also very strange.

Did you clone latest version of repo? It has different rviz config. You should look only at RED odom lines. Just disable GREEN. Arrows are perpendicular to the the direction of movement for some reason.

[laboshinl]

@drscopus

I couldn't see the constructed 3D map in rviz.

Try setting 'Decay Time' parameter to 1000 in topic 'velodyne_cloud_registered'

[drscopus]

@laboshinl Thanks! It seems works fine now!

[CansenJIANG]

Dear Leonid,

Hello, I'm a Phd student from Univ. de Bourgogne in France. And recently I found your awesome project on github using Zhang's method on SLAM. I tested your code and it was working very well. However, I have some questions on the performances of the registration results. I tested the dataset that you shared on Github (VLP16 rosbag). Here are the results that I obtain (Fig.1 is the top view, and Fig.2 is the side view) . As we know, the Zhang's method achieves the best performances among the KITTI dataset (the LOAM ranked second, while V-LOAM ranked first). But the registered clouds that I got from the dataset that you provided are bent (see Fig.2).

I would like to ask that if you are having the same problem on this testing data or not? Or do you know the reason why it happens?

Thank you very much and look forward to your reply! Cansen

Inline image 1 Inline image 2

[haoala]

Hi,

The LOAM paper defines a formula for computing curvature that normalises by the distance of the point from the origin (equation (1) in the paper). However, it seems like the curvature computed in line 378 of scanRegistration.cpp is not normalised by the distance of the point from the origin. Any idea why this might be the case?

Thanks!

[TopGunSnake]

@laboshinl When testing the loam_velodyne with real data, I noticed that the transform frames are in a non-standard coordinate system. Namely, sensor forward is mapped to +z, sensor up is mapped to +y, and sensor left is mapped to +x. Is there a reason for this, and where can I easily change it to standard cartesian?

[haoala]

@TopGunSnake I believe it's because that's the convention the paper uses (page 2 at http://www.roboticsproceedings.org/rss10/p07.pdf).

[haoala]

@laboshinl What units are distances in?

[Kailegh]

@haoala meters I think

think@laboshinl What units are distances in?

@laboshinl For the error generated in nsh_indoor_outdoor.bag is it possible to make that when it losses odometry it keeps the last pose? And I have other question, is it possible to add the help of an IMU or wheel odometry?

[laboshinl]

@Kailegh There is no error for me with current version https://github.com/laboshinl/loam_velodyne/commit/60db8c10f79bb4e2de77f6b8212b40430c2ae473

[Kailegh]

@Kailegh There is no error for me with current version 60db8c1

Thanks a lot, works perfect now I had an old version sorry. Great work man! Although this version seems to have less points, do you use a voxel grid or something like that?

Regarding the IMU, do you think there is any advantage of including one?

An important question would be what is the maximun speed this can work?

[laboshinl]

@Kailegh I have not tested it with IMU yet. https://github.com/laboshinl/loam_velodyne/issues/6

[haoala]

I believe there's a bug with laserOdometry.cpp. Lines 458-459 should really go after line 470. You want to reset the points used to build the Jacobian at the start of each new iteration, and not keep points from previous iterations.

[laboshinl]

@haoala Feel free to send pull requests with updates

[doublej317]

Can you share the original version of scanRegistration.cpp that works with HDL-32E? Thanks.

[laboshinl]

@doublej317 I was mistaken, it doesn't seem to be for HDL-32. Original sources can be found here: http://docs.ros.org/indigo/api/loam_velodyne/html/files.html

[doublej317]

@laboshinl Thank you :)