Vision-guided robotic feeding from mixed panel stacks

Need

The panels in the stack differ in size and are shifted and slightly rotated; the top layer and the panel below are the same colour, so edge separation in a 2D image is unreliable. On an overlapped panel the grasp point can land on a gap and the gripper may lift the neighbouring panel too.

Solution approach

The laser scanner on the robot arm sweeps over the stack at constant speed; profiles combined with the robot position give the point cloud. The top layer is segmented from height jumps and a rectangle is fitted to each panel; the exposed panel that does not overlap its neighbour is chosen. Its centre and angle are converted to robot coordinates, giving a grasp pose with all cups inside the panel.

Process steps

1. 01

   Scan

   The laser scanner on the robot arm sweeps over the stack; the point cloud is built.

2. 02

   Segment

   The top layer is separated from height jumps; a rectangle is fitted to each panel.

3. 03

   Grasp pose

   An overlapped candidate is skipped; the centre and angle of the exposed panel are converted to robot coordinates.

4. 04

   Pick

   The vacuum gripper descends to the panel at the computed pose and lifts it.

5. 05

   Place

   The panel is placed on the CNC infeed conveyor; the robot returns for a new scan.