[Proposal] Supporting Single LiDAR

[zusizusi]

I think the current Autoware Universe is designed to use multiple LiDARs. Therefore, single LiDAR will cause an error in the concatenate_data node. I thought that the concatenate_data node is not necessary when there is only one input point cloud, but frame transformation is also performed in the concatenate_data, and if not, an error will occur in the later processing. I could not find any other node that only performs frame transformation.

Advantages of supporting single LiDAR

• In some cases, a single LiDAR is sufficient to run Autoware on small mobile robots, etc.
• In order to make Autoware easily available to more people, it is better to be able to use Single LiDAR.
• Also, a simple system configuration will make it easier to verify Autoware's operation.

To solve this problem, I propose the following solution.

Solutions

Solution A

Change in the concatenate_data node so that if there is only one input topic, only the frame transformation is performed and the transformed point cloud is output.
I understand that a single LiDAR does not require concatenating the point cloud, but the developer only needs to change a few parameters in the sample code to make it work.

Solution B

Split the concatenate filter node into a concatenate filter and a frame transformation node, so that only the frame transformation node is executed when there is only one input point group.

Solution C

Add the function to transform frames in other filters (e.g., outlier_filter, which is one before concatenate_filter in the node diagram).
I think the output_frame parameter in node_parameters on this page should be able to be used for outliner_filter, etc.

I think solution A or solution C would be better as adding too many nodes would affect the processing time.
Please let me know if there are other solutions or other ways to use a single LiDAR.

[SakodaShintaro]

Thank you for your proposal!
AWSIM also currently uses a single LiDAR.
When using AWSIM, a large number of warnings are output to the log.

[component_container_mt-1] [WARN] [1715317124.390005323] [sensing.lidar.concatenate_data]: transformed_raw_points[/sensing/lidar/left/pointcloud] is nullptr, skipping pointcloud publish.
[component_container_mt-1] [WARN] [1715317124.390062441] [sensing.lidar.concatenate_data]: transformed_raw_points[/sensing/lidar/right/pointcloud] is nullptr, skipping pointcloud publish.
...

I also think Solution A is the best. 😄

[badai-nguyen]

@zusizusi
This is quite old discussion 😄 and I am not sure where the solution is selected for the proposal or not but I have a comment.
As my understanding, pointcloud preprocessor' node including outlier_filter is inhered from filter which already had frame transformation. The thing you should do is assign 'input_frame' and 'output_frame' where you execute the node.