noeperez
commented
3 years ago Hello Francois,
firstly, thank you very much for your interest.
You are right, the system is really sensitive to the input pointcloud and it is not easy to find a fine tuning of the parameters. This is one of the main drawbacks of the system.
The computation of the wall leaves is done by the pcl_filters node. You can try to play with the parameters pitch_max_inclination, roll_max_inclination and max_roughness of the node pcl_filters_node_1 in the file pcl_filters.launch. You could also try to change the size of the VoxelGridCovarianceFilter in the file rrt_filter_params.yaml. Your general input pointcloud should be dense enough to obtain good results of this filtering.
On the other hand this sound unusual to me: "-I use Octomap to read LIDAR data and publish a PointCloud2 map under "odom_combined" frame (equal to "odom" frame) via /octomap_point_cloud_centers topic I give this topic to indires_navigation."
The system is waiting to receive a pointcloud map from a mapping system in a global frame (for instance: /map or /world), not in a odometry frame. The general TF tree should have a general structure similar to: map frame (poincloud frame) -> odom frame -> robot_frame -> robot_sensors, wheels, etc. I'm not sure if your structure could lead the system to fail.
My guess, it's that move_base is failing because of any computation with the pointcloud is not giving the expected results or even the TF structure.
I hope these comments can help you. Thank you Francois.
Best regards, Noé
El jue, 6 may 2021 a las 11:50, FrancoisTMM @.***>) escribió:
Hello ! Thank you for this repo which looks great for 3D ground navigation.
I am an ingineering student, and I try to integrate your work for a project. I have some issues, and I need help to solve them.
First, this is my system :
-I run it under Ubuntu 18.04 with ros-melodic
-I work in a simulation (webots). I use a diff drive (4 wheels) robot named Sylvester, which perceive is environment from a LIDAR (Velodyne VLP-16). He reads command into /cmd_vel topic and produces LIDAR data via /sylvester/velodyne_pointcloud topic. LIDAR frame is "velodyne".
This is a picture of a part of my simulation : [image: simulation_idires_issue] https://user-images.githubusercontent.com/45355784/117269595-b4e20b00-ae58-11eb-9c61-759b56c37ce0.png
-I use Octomap to read LIDAR data and publish a PointCloud2 map under "odom_combined" frame (equal to "odom" frame) via /octomap_point_cloud_centers topic I give this topic to indires_navigation.
An example of point cloud map that I can perform [image: map_indires_issue] https://user-images.githubusercontent.com/45355784/117271444-7cdbc780-ae5a-11eb-9f44-e0d6e82a93ae.png
This is a picture of my TF tree [image: rqt_graph_indires_issue] https://user-images.githubusercontent.com/45355784/117270306-5d906a80-ae59-11eb-93e8-ac1eb4b249b3.png
-Now, I want use indires_navigation to explore robot's environment.
But, that is why I am here for now... I have some issues...
The one I have the most of time is when I try to use the exploration mode. After cumputation time RRT found wall leaves every where, ans some times, some frontier: (red = walls / yellow = frontier, as your default color configuration) [image: wall_leaf_indires_issue] https://user-images.githubusercontent.com/45355784/117275142-0640c900-ae5e-11eb-8c95-10fc76129128.png
this is the features point cloud (which I think is used to compute wall and frontiers leaves): [image: features_PC_indires_issue] https://user-images.githubusercontent.com/45355784/117275663-8d8e3c80-ae5e-11eb-9dfb-1bf54f407f22.png
the RRT sample pointcloud: [image: rrt_sample_PC_indires_issue] https://user-images.githubusercontent.com/45355784/117275706-9aab2b80-ae5e-11eb-93ac-db0c4c3146cb.png
and the local control pointcloud: [image: local_control_PC_indires_issue] https://user-images.githubusercontent.com/45355784/117275794-aeef2880-ae5e-11eb-98dc-c942f5e1c75d.png
I tried to modify a lot of params to have a good behavior into several files : 3dfeatures_params.yaml / rrt_costmap_params.yaml / navigation_params.yaml (the one from adapted_move_base)... But this result is the best that I could perform. I tried to change frontier treshold, pcl method to compute pointclouds, different leaf size or minimal number of point... Futher more, after computing, a goal is given to controller, and adapted_move_base crash, after this error :
terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
[adapted_move_base_node-6] process has died [pid 6161, exit code -6, cmd /sylvester_indires_workspace/devel/lib/adapted_move_base/adapted_move_base_node __name:=adapted_move_base_node __log:=/root/.ros/log/a4996916-ae4c-11eb-80d9-d89ef338b879/adapted_move_base_node-6.log].
I don't figure what could turn wrong into my configuration, but I guess that should come from a wrong params tunning that don't fit correctly with my octomap point cloud. After some days of investigation, I finally think I need your help to solve that.
Best regards ! François
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/noeperez/indires_navigation/issues/11, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADUK36DIPM6ESTCIJBXBN5TTMJQ47ANCNFSM44GWKWNA .
FrancoisTMM
Hello ! Thank you for this repo which looks great for 3D ground navigation.
I am an ingineering student, and I try to integrate your work for a project. I have some issues, and I need help to solve them.
First, this is my system :
-I run it under Ubuntu 18.04 with ros-melodic
-I work in a simulation (webots). I use a diff drive (4 wheels) robot named Sylvester, which perceive is environment from a LIDAR (Velodyne VLP-16). He reads command into /cmd_vel topic and produces LIDAR data via /sylvester/velodyne_pointcloud topic. LIDAR frame is "velodyne".
This is a picture of a part of my simulation :
-I use Octomap to read LIDAR data and publish a PointCloud2 map under "odom_combined" frame (equal to "odom" frame) via /octomap_point_cloud_centers topic I give this topic to indires_navigation.
An example of point cloud map that I can perform
This is a picture of my TF tree
-Now, I want use indires_navigation to explore robot's environment.
But, that is why I am here for now... I have some issues...
The one I have the most of time is when I try to use the exploration mode. After cumputation time RRT found wall leaves every where, ans some times, some frontier: (red = walls / yellow = frontier, as your default color configuration)
this is the features point cloud (which I think is used to compute wall and frontiers leaves):
the RRT sample pointcloud:
and the local control pointcloud:
I tried to modify a lot of params to have a good behavior into several files : 3dfeatures_params.yaml / rrt_costmap_params.yaml / navigation_params.yaml (the one from adapted_move_base)... But this result is the best that I could perform. I tried to change frontier treshold, pcl method to compute pointclouds, different leaf size or minimal number of point... Futher more, after computing, a goal is given to controller, and adapted_move_base crash, after this error :
I don't figure what could turn wrong into my configuration, but I guess that should come from a wrong params tunning that don't fit correctly with my octomap point cloud. After some days of investigation, I finally think I need your help to solve that.
Best regards ! François