Mapping based lidar - lidar calibration 'lost' frames

[Acuno41]

Greetings,

My system:

• Ubuntu 22.04
• Ros2 Humble
• Fast DDS

I was trying to do mapping based lidar-lidar calibration tutorial and I faced some issues that i want to ask.

At first I run the sensor_calibration_manager and set the project "rdv" and calibrator "mapping_based_lidar_lidar_calibrator" with default parameters then download and run rosbag which shared in tutorial file with half speed and everything goes well. But when i run ros2 service call /stop_mapping std_srvs/srv/Empty command, ı faced the error below.

When i tracked the error problem looks caused by this lines. Then I updated that line like below and that problem gone.(I checked the updated result and it was desired Dict[str, Dict[str, np.array]] format)

        result = {
            self.sensor_kit_frame:
                dict(
                    zip(
                        self.calibration_base_lidar_frames,
                        sensor_kit_to_calibration_lidar_transforms,
                    )
                )

        }

My first question is that fix OK or should it be something different ?

And my second question is when I updated the code above I run and stop mapping successfully but cant get any calibrations due to 'lost' frames. The terminal output below. I tried to run bag much more slow and fast, changed some parameters and nothing works.
How can I able to create successfull calibration with tutorial data ? Do you have any advice ?

Best regards.

[knzo25]

@Acuno41
Hi, thank you for your report.
Regarding the first issue, it is probably a bug on our side. I will try to fix it as soon as possible, but can not really promise any date. if you can contribute with a PR, we would highly appreciate it.

Regarding the second issue, lost frames are usually caused because the lidar was too sparse, the vehicle was moving to fast, or there were not good enough features for registration.
Since this happens in the sample rosbag, the cause would be that some data samples are simply being lost. We do not have experience with Fast DDS (we use cyclone DDS) but the cause may be there. Additionally, I would recommend trying with a -r 0.1 replay rate just to be sure.

[Acuno41]

Thanks for the advices @knzo25.

Sure I can create PR about it but I'am also a little bit busy for next couple days. I saw couple of other python files have the same issue. I will also fix them.

For your second answer, I tried your suggestions and problem was with the DDS when I changed Fast DDS to cyclone DDS calibration works fine no frame lost. But honestly, I don't realy understand the difference behind the DDS vendors which creates huge impact.

I also remembered when I installed the repository via Standalone installation (for non-autoware users) but repo threw error that says about some packages autoware_utils. It may be autoware_utils in vcs calibration_tools_standalone.repos outdated. FYI.

[knzo25]

@Acuno41

Good to know you found the cause !

While DDS allows us to tune the system to meet our needs in terms of data volume, etc, it also makes it very difficult to use, and we often struggle with that on new projects.

Regarding the standalone installation, the cause is that autoware is being re-engineered at the moment, and the CI/CD does not make it mandatory to make sure this repository compiles. Although I should make CI/CD's here to know when it breaks, the dependencies over autoware are so loose that it had not caused an issue until very recently. I am even tempted to just copy the one or two classes that I need, which would make this repository quite lighter 😅