[TASK] Pipeline Following

[gardeg]

Description of task

Create a pipeline following code. This will include Research of CNN or Computer-vision Algorithm. Training models for testing and potential final product. Testing and optimization.

Suggested Workflow

• CNN research - [RESEARCH] CNN training vortex-auv#601
• Create prototype of CNN line following, preferably both Detection and classification - [TASK] Prototype CNN vortex-auv#610
• Prototype of CNN segmentation already exist but can potentially improve. - https://huggingface.co/vortexntnu/pipeline_segmentation/tree/main
• Decide final approach ether CNN or Classic computer vision (done before new year, latest early January)
• Create final product setup
• Testing and optimization

Specifications

• The pipeline following algorithm should be robust and run with a low latency, so that
  inaccuracies can be filtered out by new predictions before a divergence occurs.
• Ensure the AUV is positioned correctly to detect all ArUco markers along the
  pipeline.

Contacts

• @gardeg

Code Quality

• Every function in header files are documented (inputs/returns/exceptions)
• The project has automated tests that cover MOST of the functions and branches in functions (pytest/gtest)
• The code is documented on the wiki (provide link)

[jorgenfj]

Have you thought about using a regression model instead of a simple classification model? This way we could predict a continuous angle instead of discrete values like RIGHT,LEFT or FORWARD.

[kluge7]

1. Make a custom ROS message that will be sent in vortex-msgs (
2. Integrate it and implement it in line fitting.

[gardeg]

need to make all of the line following logic as a function that uses the filtered line.

take the line and make waypoints that the line can follow, it needs to have control of the rotation of the drone relative to the drone. and fetch the relative position of the drone and make a new waypoint based on the position.

the filter is a function the gets fetched