feat: add ground_segmentation_evaluator

[nanoshimarobot]

Types of PR

• New Features
• Upgrade of existing features
• Bugfix

Description

Add ground segmentation evaluator scenario.
This evaluator has two types of ground truth data.

• annotated pcd
  *.pcd.bin files located in t4_dataset/sample/data/LIDAR_CONCAT/ and the data has semantic label, provided by #107.
• annotated rosbag
  The rosbag data in the input_bag contains annotated pointcloud data, captured from sources such as AWSIM.

How to review this PR

Please download this driving_log_replayer_data for ground_segmentaion evaluation: TIERIV_INTERNAL_LINK
That data has only annotated pcd data, so please also check this annotated rosbag data: TIERIV_INTERNAL_LINK

Please download this driving_log_replayer_data for ground_segmentaion evaluation : TIERIV_INTERNAL_LINK
There are separate scenario.yaml and input_bag files for each mode, as shown in the image below:


Therefore, for testing, please modify each name to 'scenario.yaml' and 'input_bag/' before running the tests.

Others

• When running ground_segmentation in 'annotated_rosbag' mode, it is necessary to modify the sensing module of autoware.universe as follows:
  https://github.com/autowarefoundation/autoware.universe/blob/main/sensing/autoware_pointcloud_preprocessor/src/filter.cpp#L383-L390

       if (utils::is_data_layout_compatible_with_point_xyzi(*cloud)) {
        RCLCPP_ERROR(
          get_logger(),
          "The pointcloud layout is compatible with PointXYZI. You may be using legacy "
          "code/data");
      }
  
      // return; <- comment out!
  }
  

• When running ground_segmentation in 'annotated_pcd' mode, it is necessary to reduce the rosbag play rate because the evaluation process takes longer than the ground_segmentation process in this mode.
  dlr simulation run -p ground_segmentation --launch_args play_rate:=0.1

[badai-nguyen]

@nanoshimarobot
Since Method: annotated_pcd case will use t4_dataset input format, I think we needed to add ground_segmentation in here

driving_log_replayer/driving_log_replayer_cli/simulation/run.py

Line 17 in 7c04a66

USE_T4_DATASET = ("perception", "obstacle_segmentation", "perception_2d", "traffic_light")

[badai-nguyen]

could you test with pilot-auto?

Suggested change

	VehicleModel: jpntaxi
	VehicleModel: lexus

[ktro2828]

Because aip_xx1 and jpntaxi are TIER IV internal models, it is better to set them to sample_sensor_kit and sample_vehicle like other sample scenarios for perception tasks in driving_log_replayer.

driving_log_replayer/sample/perception/scenario.yaml

Lines 4 to 5 in a91986e

	SensorModel: sample_sensor_kit
	VehicleModel: sample_vehicle

[ktro2828]

[Q] Are these comment out intended?

[nanoshimarobot]

No, those comments were unintentional. I have fixed them.

[sonarcloud]

Quality Gate passed

Issues
3 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.4% Duplication on New Code

See analysis details on SonarCloud