Cutting leeks accurately with good repeatability (the name of the game) using automation is difficult because of the sheer variation and randomness of leeks - or any vegetable, for that matter. Using 3D and AI we try to emulate what a human would do.
We started out by just looking at the shape, easy right? It's much more complex than that. We've found three main scenarios which account for better than 95% of what is seen by our 3D camera so the system has to use each technique, then score it accordingly. The highest score wins and with that, the leek is cut at the right point and at the right angle.
Once the cut solution has been defined (in a fraction of a second), the coordinates are sent to a servo which moves a water blade to cut at the right point. The result is nicely cut leaks with minimal waste, with a longer shelf life.
We can apply the same process to any kind of fruit or vegetable.
The following images show the basis behind each decision.
Here's a video of the result, from the camera's perspective!