Revision history [back]

There is a Discourse thread regarding this issue:

https://discourse.ros.org/t/image-and-pointcloud2-performance-on-ros2/5391

In short: The performance on large messages like PointCloud2 seems to be much worse in ROS2 than in ROS1. Since transferring large point clouds is a common use case, this is unacceptable, and I expect this to be fixed at some point. This is the core issue behind your problem, so it should be addressed first.

That being said, it sounds like a great idea to offer point cloud compression! However, before a message is added to the core ROS message packages like sensor_msgs, there should be at least one (but better several) implementations that use the proposed message for a longer time. So the first step would be for you to create that message in your own code and provide an implementation. If this is successful and there is wide-spread interest, one can begin the process of adding the message to sensor_msgs. If you want to go down this road, let me suggest Google Draco for point cloud compression:

https://github.com/google/draco

I haven't used it myself yet, but it looks extremely intriguing.