LiDAR Detection (Issue #368)

[MtGuerenS]

Description

Implemented a lidar detection node that combines the MMDetection3D and the ComplexYOLOv4 models for the Object classification (Lidar) Issue.

Usage

To test the new feature do the following:

1. Click here and download the complex_yolov4_mse_loss.pth file
2. Move complex_yolov4_mse_loss.pth into the directory navigator/data/perception/checkpoints/complex_yolov4_mse_loss.pth
3. cd into /navigator
4. uncomment # *object3d_launch_entities, in launches/launch.perception.py
5. run docker compose run navigator_carla
6. run colcon build --symlink-install --packages-select lidar_detection && source install/setup.bash
7. run ros2 launch launches/launch.vehicle.py
8. open new ternminal with pwd in /navigator
9. run docker compose run navigator_carla
10. run ros2 bag play /bags/2024/bag_2024-02-09_Demo/ or other bag

[danielv012]

Everything looks awesome. Just a few comments I left

[danielv012]

What's with all the changes to the default rviz config?

[danielv012]

Was this a rebase?

[MtGuerenS]

I added a group for object detection/tracking, and the rest of the changes are just the position I had the camera at. Here is the most important part of the changes:

- Class: rviz_common/Group
      Displays:
        - Class: rviz_default_plugins/MarkerArray
          Enabled: true
          Name: Detected
          Namespaces:
            object_bounding_box: true
            object_tag: true
          Topic:
            Depth: 5
            Durability Policy: Volatile
            History Policy: Keep Last
            Reliability Policy: Reliable
            Value: /viz/detected/objects3d
          Value: true
        - Class: rviz_default_plugins/MarkerArray
          Enabled: false
          Name: Tracked
          Namespaces:
            {}
          Topic:
            Depth: 5
            Durability Policy: Volatile
            History Policy: Keep Last
            Reliability Policy: Reliable
            Value: /viz/tracked/objects3d
          Value: false
      Enabled: true
      Name: 3D Objects

[ragibarnab]

Hi Gueren. Please don't hardcode the paths to model configs/weights, change it so they pass them through ros params. There's a reason I used MMDetection. That is, you can take any state-of-the-art model (a lot of them better than ComplexYOLO) from MMDetection package and use it for inference. By hardcoding the configs/weights the key functionality of my package is lost.

Thanks,
Rae

P.S. in the future just communicate with others if you are going to edit and remove their code. Communication is key, y'know? I'm a nice guy, I promise, just reach me out on teams. Like really, our packages should've been split instead of combined the way it is to prevent any issues.