Students: Sjoerd Kaptein, Koen Mourits, Marc Groeneveld. (2017)
Scindapsus Potting
Background
This project is part of the minor Smart Manufacturing and Robotics located at The Hague University in Delft. The project has been initiated by Demokwekerij Westland to research the automation of cutting and potting the Scindapsus plant breeded by W. Kap. The project continues with a prototype made by former Smart Manufacturing and Robotics students. Their assignment was to cut a string of the Scindapsus plant into stems of 1,5 to 2 cm that contain a root to plant into the ground.
Greenhouse W.Kap has been specialized in growing Scindapsus plants (Epipremnum) for years. Through years of experience W.Kap has grown into one of the leading Scindapsus greenhouse. W.Kap is reliable, innovative and distinctive!
Because of this W.Kap is in a constant search for improvements for his company and automation of the potting process is a logical next step in this process. The current state of this process is that four employees are cutting the Scindapsus strings in little pieces to pot, they produce 200 stems per hour in total. At the same time a machine fills pots with soil after which they are placed on a table of 1,7 by 3,6 m with a tool that can pick up one row ate a time. As soon as all the pots are on the table the workers will pot the freshly cut stems by hand taking up another hour. Adding this up makes it that all workers can effectively cut and pot 1000 stems per hour.
Project Assignment
As mentioned in the background information the assignment continues with a prototype made by former Smart Manufacturing and Robotics students. But now the assignment of this project is to design and make a system that can be used to pot those Scindapsus plants’ stems into a pot. The pots are placed on a table
in a random order. The system should be able to identify the pots and their location. The cut stems are placed in an array and the system can pick the stems from this array. Using a robotic arm and machine vision cameras, the robotic arm must be able to recognize the orientation of the cut stems and pot them autonomously in the right orientation. The orientation in which the plants will be potted is very important because the leaves of every stem must point in the same direction. Since the tables currently used by the greenhouse are too big for the robotic arms located at the HHS, a scale prototype of the system will be made during this project. The kawasaki FW03N robotic arm will be used to demonstrate the potting process.
Problems
The camera’s used for the stem and pot vision are wide angle lens cameras, because of these lenses there is distortion in the picture. Therefore calibrating Halcon was a challenge and had to be done by hand. Those wide angle lens also made the recognising of the pots harder, because the pots in the middle were more circular than those on the sides. But the biggest challenge in the pot vision was the fact that pots were touching each other and sometimes the soil was even overlapping which made it look like two pots were connected by a bridge of soil. This makes it very hard to recognize every single pot as a separated one.
The stem vision was quite a challenge as well. After some picture tests with just the grippers it was clear that a good background was needed so a background board was made and put behind the grippers. Having a white background created good contrast. Using different methods the leaf was cut off the picture
so only the stem was visible. Using the center of this remaining stem the angle of the leaf was determined.
One of the first problems we solved was the holding/gripping of the stems without damaging the them. The hardest part about this is that plants are never the same the thickness of the stem varies and many of them are corked. The leaves could be in imaginable position, in short: Handling plants is very hard to automate.
Results
The system can recognize where the pots are on the conveyor belt and the orientation of the leaf. This is being achieved by two separated Vision camera setups working together to create a set of coordinates including the rotation of the leaf. First the pots are being recognized by the Pot Vision program, the centre and size of the pots are calculated and saved in an array waiting for the additional information from the Stem Vision program. The robot picks up a stem and holds it in front of the Stem Vision camera. The Robot sends a signal to the server and the Stem Vision program starts running. The orientation of the leaf is determined and combined with the information gathered from the Pot Vision. The coordinates of the pots and orientation of the leaf are being transformed into the
coordinate system used by the Kawasaki robot. The robot moves to the coordinates that are given, during that movement the stem is rotated in the right direction.
To get the conveyor belt to move and stop at the right time a programmable logic controller is used. Using digital inputs and outputs the PLC is connected with the Kawasaki robot
Conclusion
The main goal of this project was to build a prototype that could pot cut stems into three different sized pots, with the leaves orientated in the same direction. This goal is accomplished. However, a lot of improvements have to be made in order to make this robot robust enough to implement inside a green house. The End of Arm Tool can be optimised so that it can pick up multiple stems at once. The way of detecting the pots is too complicated it is recommended to get the pots in fixed positions. To match the required speed and reach a different robot is needed.