feat: add tool for open-set detection using grounding dino

[rachwalk]

Purpose

Adds Grounding Dino Tooling for distance estimation. Grounding Dino is used to extract the regions of the camera POV classified as certain objects. These regions can then be extracted from the depth channel, and after calibration, used to estimate distance from camera to them.

Proposed Changes

• Adds two tools: GetDetectionTool, GetDistanceToObjectsTool.
• Defines API for the grounding_dino extension

Issues

#194

Testing

Tested using following script:

import rclpy
from rai_grounding_dino.tools import GetDetectionTool

from rai.node import RaiBaseNode

if __name__ == "__main__":
    rclpy.init()
    node=RaiBaseNode(node_name="unique_node")
    x = GetDetectionTool(node)._run(
        camera_topic="/camera/camera/color/image_raw",
        object_names=["chair"],
    )
    print(f"Detected: {x}")
    x = GetDistanceToObjectsTool(node)._run(
        camera_topic="/camera/camera/color/image_raw",
        depth_topic="/camera/camera/depth/image_raw_rect",
        object_names=["chair"],
    )
    print(f"Detected: {x}")

In second terminal run:

ros2 launch rai_grounding_dino gdino_launch.xml

Camera topics must be available and of the same resolution. The detection went through correctly.

[adamdbrw]

Could you include the test in the commit, and document the new tool as well for users and developers? Does it require additional dependencies?

[maciejmajek]

Thanks. I can confirm that the feature works as expected. Please take a look at the comments.
Regarding the {... or 'Nothing'}, decide yourself whether handling it that way or using if measurements[] is better