Gap in the point cloud on Velodyne sensor

[mebasoglu]

Description

When I use Nebula with VLP-16 on main branch, I saw a gap on the point cloud. The gap is flickering. Here is a video of the point cloud:

nebula-gap.mp4

Expected behavior

We should have a full point cloud without any gaps.

Actual behavior

There is gap which is flickering on the point cloud.

Steps to reproduce

Run the launch file for Velodyne on the main branch:

ros2 launch nebula_ros velodyne_launch_all_hw.xml sensor_model:=VLP16 sensor_ip:=192.168.1.208 data_port:=2368

And on the RViz2 screen the point cloud:

Additional context

I can provide a pcap file that I recorded from the sensor if it is needed.

[mebasoglu]

I think this bug is related to the scan_phase parameter. In here I have added a feature called "dynamic scan phase" and I don't see a gap in the point cloud with it. It works like this:

1. When a UDP packet is received, the current system clock time is rounded up to the next second and a "first publish time" value is created if the UDP packet is the first received one. And until that first publish time comes, packets are skipped.
   https://github.com/leo-drive/nebula/blob/d53e0ff5a915f90466daa025ed23e7ac6e6d5464/nebula_hw_interfaces/src/nebula_velodyne_hw_interfaces/velodyne_hw_interface.cpp#L87

2. After reaching first publish time, the azimuth angle of the most recent received packet is published to "/scan_phase" topic. This publishing happens only once.

3. The decoder node subscribes to "/scan_phase" and sets the parameter. Instead of using scan_phase from parameter, this new value is used while decoding the packets.
   https://github.com/leo-drive/nebula/blob/memin/dev/dynamic-scan-phase/nebula_ros/src/velodyne/velodyne_decoder_ros_wrapper.cpp#L93

This feature allows multiple driver instances publish PointCloud2 messages as close as possible so that the concatenate node won't miss any of the PointCloud2 messages. Here is the rqt_graph screenshot of raw point clouds:

With this dynamic scan phase change feature, the gap issue was also solved.

What are your opinions about this? @drwnz @amc-nu

Thank you.

Edit: Upon further investigation I realized that this didn't solve the issue. The gap still exists. :(

[mebasoglu]

On the get_pointcloud() method in vlp16_decoder I changed the angle type from radians to degrees like the previous awf_velodyne driver and it worked.

Instead of these:

nebula/nebula_decoders/src/nebula_decoders_velodyne/decoders/vlp16_decoder.cpp

Lines 63 to 64 in f82a6f8

	auto current_azimuth = scan_pc_->points.back().azimuth;
	auto phase_diff = static_cast<size_t>(angles::to_degrees(2*M_PI + current_azimuth - phase)) % 360;

I added:

auto current_azimuth_deg = angles::to_degrees(scan_pc_->points.back().azimuth);
auto phase_diff_deg =
      static_cast<size_t>(360 + current_azimuth_deg - sensor_configuration_->scan_phase) % 360;

And the new while loop:

while (phase_diff_deg <= 180 && !scan_pc_->points.empty()) {
      overflow_pc_->points.push_back(scan_pc_->points.back());
      scan_pc_->points.pop_back();
      
      // current_azimuth = scan_pc_->points.back().azimuth;
      // phase_diff = static_cast<size_t>(angles::to_degrees(2*M_PI + current_azimuth - phase)) % 360;

      current_azimuth_deg = angles::to_degrees(scan_pc_->points.back().azimuth);
      phase_diff_deg =
        static_cast<size_t>(360 + current_azimuth_deg - sensor_configuration_->scan_phase) % 360;
    }

And now the gap is gone. I have only tested for VLP-16.

[amc-nu]

Fixed in #40
and
#50

[mebasoglu]

Hello, I still face this issue with most recent commits on main branch.

[drwnz]

@mebasoglu thank you, do you have a rosbag with packets topic you can share?
I have tested on our devices and there doesn't seem to be an issue, so if you have some data we can look at, that would be helpful!

[mebasoglu]

Hello @drwnz , while working on the vehicle I realized that it was my mistake to set scan phase properly. After configuring it correctly, there is no problem. I apologize for taking your time. Thank you.