Assumptions : we have a robot arm ROS package. using an 3d camera, like Intel Realsense D455.(or any other). we can add a line laser scanner.
Arm and camera already calibrated.
Task : need to identify the glue seams and plan the path for them, with sub-mm precision.
Work testing : we place a part on the table, in the visual field of the camera. (From segmentation) it identifies the plane of the table, and the part itself. (whatever the shape)
Part. A. It sees and displays the glue lines on the screen/gui/rviz/where ever. (Mostly corners and angles, sometimes rounds and curves)
Part B. If it needs to scan more, it produces the path planning for scanning. If not it produces the glue path lines.
We can use a .step/CAD file to define the lines in a gui previously, (and do 6dof with ViSP) but it would be best to identify them from the pcl and just confirm them by hand. Later on, will try to use an NN to self- identify lines that need to be glued.
Attached and example of part as reference, but imagine that parts are any model,type, size in the view of the camera.