lp-research / lpslam_node Goto Github PK

WARN: Recipe 'tbb/2020.1' is deprecated in favor of 'onetbb'. Please, consider changing your requirements.
ERROR: Conflict in protobuf/3.9.1:
'protobuf/3.9.1' requires 'zlib/1.2.12' while 'boost/1.72.0' requires 'zlib/1.2.13'.
To fix this conflict you need to override the package 'zlib' in your root package.
CMake Error at lpslam/cmake/conan.cmake:402 (message):
Conan install failed='1'
Call Stack (most recent call first):
lpslam/cmake/conan.cmake:497 (conan_cmake_install)
lpslam/CMakeLists.txt:101 (conan_cmake_run)

New to this, not sure how should I approach this problem. Any help will be appreciated. Thank you.

So far we don't have any public documentation linked to this project, not even build instructions. Need to add. Potentially we have something internally at LP, will check.

Hi! Listing the items more or less required for first Nav2 integration (will be updated during the ticket lifetime).

• Set nput/output tunes (configs and topics) as input ROS-parameters (#9)
• Switch to time synchronizes sensors subscription model (#10)
• Back-port of "initial_pose" feature from OpenVSLAM (locally done at https://github.com/AlexeyMerzlyakov/lpslam_node/tree/initial_pose, https://github.com/AlexeyMerzlyakov/lpslam/tree/initial_pose, https://github.com/AlexeyMerzlyakov/lp-openvslam/tree/initial_pose branches)
• Support camera calibration parameters to read from input topic (#11, #13)
• Add the support for RGB-D cameras
• Make an option for "lpslam_ros" to work without "lpslam" module which is unnecessary for ROS2 (#14)
• Fix crashes in localization mode (#7)

Last output: [info] Occupancy Map: Computed needed grid size to..
[from openvslam/module/occupancy_map_exporter.cc@compute_parameters()]

Running the node in debug mode:

ros2 run --prefix 'gdb -ex run --args' lpslam lpslam_node --ros-args --log-level lpslam_node:=DEBUG --log-level rcl:=INFO -p lpslam_config:="PATH_TO_CONFIG.json" -p use_sim_time:=True -p laserscan_frame_id:=LASER_FRAME -p write_lpslam_log:=True -r left_image_raw:=RIGHT_TOPIC -r right_image_raw:=LEFT_TOPIC

gdb says, it comes from get_laser_rays_in_world()

Segmentation fault in openvslam::data::laser_landmark::get_laser_rays_in_world() const () 
from ...install/lpslam/lib/libopenvslam.so

Following the functions involved up to lpslam_node:

|___ openvslam/data/laser_landmark.cc@get_laser_rays_in_world()
|___ openvslam/laser/laser_scanner_2d.cc@get_laser_rays_in_world()
|
|___ openvslam/module/occupancy_map_exporter.cc@map_export()
|___ openvslam/publish/map_publisher.cc@occupancy_map_export()
|
|___ lpslam/src/InterfaceImpl/LpSlamManager.cpp@mappingGetMapRaw()
|___ lpslam/src/Trackers/OpenVSLAMStereoTracker.cpp@mappingGetMapRaw()
|___ lpslam/src/Manager/SlamManager.cpp@mappingGetMapRaw()
|
|___ lpslam_node/src/LpSlamNode.cpp@publishOccMap()

Apparently, the problem happens when LpSlamNode tries to publish the occupancy map and there is a map size mismatch at some point or a problem with laser data.
If it's just an initialization problem, adding an exception to let the node "try to publish the map" would let it carry on but this needs more investigation.

Debugger output,

Program received signal SIGABRT, Aborted.
__GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
50	../sysdeps/unix/sysv/linux/raise.c: No such file or directory.

@AlexeyMerzlyakov figured out that the crash is caused by spdlog version conflict. In our case spdlog/1.8.0 requested by lpslam/conan vs system spdlog/1.5.0 vs OpenVSLAM 3rd/spdlog.
So when doing manual changes to force all three to use the same spdlog version, the node works just fine!

Hello LP-RESEARCH development team,

I just found this repository a few minutes ago, and I'm currently confirming your codes.

And I found (probably unintentional) typo lines (at current HEAD commit) shown as below.

I know this is just my quibble on insignificant detail, but I hope this will be of some help.

In the lines shown below , "lpslam or LPSLAM" is used.

In the lines shown below , "LPSPAM" is used.

Thanks,

For the USE_OPENVSLAM_DIRECTLY=OFF node, lpslam module is doing the preprocessing internally in lpslam/OpenVSLAMStereoTracker.cpp and not with OpenVSLAM but it also gives the option to use OpenVSLAM rectifier depending on whether StereoRectifier.model is set in the configuration file or not as specified in openvslam/system.cc.
But this doesn't work because the value is not being parsed properly and will always be False.

On the other hand, stereo rectification is needed to improve the stereo matching and the map quality.

Issue raised by @SteveMacenski (afaik):
At the moment the occupancy grid is created from laser scan data aligned to the VSLAM coordinate system. Laser scan segments are connected to camera key frames and (re)arranged correspondingly.

This should be more flexible. There are ways to create an occupany grid from camera data. Exploration of how to actually do that is another issue. This issue is for creating some kind of flexible interface.

Comment from @ThomasHauth:
Actually the code involved for the occ grid generation is not that much right now. So I don't know if it "deserve" its own library right now. Its all contained in this file:
https://github.com/lp-research/openvslam/blob/feature/lprmods/src/openvslam/module/occupancy_map_exporter.cc

This is a broad issue. Will add more details here later.

Some of the stability issues are shown here:
https://vimeo.com/553403043/228cecde7d