Point-cloud-based methods in Open3d-ML

[sebbyjp]

Create a new subclass of SensoryAgent that returns a Sample subclass representing the objects and volume bounding boxes.

Steps:

• 1. Experiment with pre-trained models in Open3D-ML
• 2. Define appropriate Sample subclass that contains 3d volume bounding box information.
• 3. Load weights from select object-detection and semantic-segmentation models from Open3D-ML and upload to HuggingFace
• 4. Implement SensoryAgent interface.
• 5. Add unit tests and create Gradio App with examples for qualitative inspection.

[emekaokoli19]

I'll take this

@sebbyjp Just to be clear, the SensoryAgent subclass implements the sense() function which takes the PointCloudSample type as an input parameter, uses an object detection model from open3d-ml to get the bounding box, and returns it as a SensorReadingSample type?

[sebbyjp]

Yes, sir! sense() is an alias for act() in this case.