More and more harvesting robots are being used around the world, especially for expensive fruit, mainly to replace missing workers. For example, robots choose Strawberries, tomatoes, apples, peppers or kiwis.
Announcement
Basically, there are two different approaches: the robot shakes the tree or bush and then picks up the falling fruit, or – especially in the case of sensitive fruit – the machine approaches the object to be picked up with a gripper and picks it up. However, some fruits, such as raspberries, are extremely delicate and are therefore easily destroyed by the robotic grippers.
Training on artificial raspberries
Kai Junge from EPFL Lausanne and colleagues have now trained a robot to solve this problem with the help of a ‘physical twin’ in the form of an artificial raspberry equipped with sensors.
Recommended editorial content
With your consent, an external YouTube video (Google Ireland Limited) will be uploaded here.
Always upload YouTube videos
Robots train in picking artificial raspberries
The artificial raspberry consists of a 3D printed outer skin filled with liquid silicone. The external pressure force when gripping the fruit is measured by a fluidic sensor. The sensor raspberry is attached with a magnet to a flexible artificial tendril, which is also 3D printed. The distance between the two halves of the magnet defines how large the maximum pulling force is to detach the sensory raspberry from the tendril.
Junge and colleagues first asked people to choose this sensory raspberry, tweaking parameters and recording strength profiles. They then used it to optimize the parameters for gripper control. The robotic gripper, which sits on a standard robotic arm, has a Raspi camera that first locates and approaches a raspberry. Then two soft silicone fingers close around the berry and the robot starts pulling until the fruit comes off. As soon as this happens, the inspector should ensure that the lateral force on the berry is eased as much as possible, without letting go of the fruit. The authors explain the technical details in their paper.
Announcement
The robot was having trouble finding raspberries
With the physical twin training, the authors above all want to make costly and time-consuming field tests superfluous. Because if such a system works in the laboratory, it is still a long way from being able to use it in practice. Indeed, the hardware and software must first be adapted to the complex conditions of use. However, this is usually only possible if there are also fruits ready to be harvested for training – this period is however limited.
With the Raspberry model, the researchers then looked at a whole range of parameters – stiffness, peel force, etc. – in order to make the controller as universally applicable as possible. However, the practical test showed that the machine picked 80% of the ripe fruit correctly. However, the robot had trouble finding raspberries hidden by tendrils, leaves or fruit and in some cases failed to muster enough strength.
A similar project shows how long the transition from lab to practice can be: Start-up Fieldwork Robotics, founded by the University of Plymouth’s agricultural robotics group, has had two copies of a raspberry picking robot in practical use.
According to the Guardian At that time, the four-armed robots were capable of picking one kilogram of fruit per hour, which is expected to be increased to four kilograms per hour in the next stage by 2023. The first version of the robots, under development since 2018, took about a minute per pod.
(wst)
