Harvesting fruits and vegetables is a complex task worth to be fully automated with robotic systems. It involves several precision tasks that have to be performed with accuracy and the appropriate amount of force. Classical mechanical grippers, due to the complex control and stiffness, cannot always be used to harvest fruits and vegetables. Instead, the use of soft materials could provide a visible advancement. In this work, we propose a soft, sensorized gripper for harvesting applications. The sensing is performed by tracking a set of markers integrated into the soft part of the gripper. Different machine learning-based approaches have been used to map the markers’ position and dimensions into forces in order to perform a close-loop control of the gripper. Results show that force can be measured with an error of 2.6% in a range from 0 to 4 N. The gripper was integrated into a robotic arm having an external vision system used to detect plants and fruits (strawberries in our case scenario). As a proof of concept, we evaluated the performance of the robotic system in a laboratory scenario. Plant and fruit identification reached a positive rate of 98.2% and 92.4%, respectively, while the correct picking of the fruits, by removing it from the stalk without a direct cut, achieved an 82% of successful rate.
A soft, sensorized gripper for delicate harvesting of small fruits
Francesco Visentin
;Fabio Castellini;Riccardo Muradore
2023-01-01
Abstract
Harvesting fruits and vegetables is a complex task worth to be fully automated with robotic systems. It involves several precision tasks that have to be performed with accuracy and the appropriate amount of force. Classical mechanical grippers, due to the complex control and stiffness, cannot always be used to harvest fruits and vegetables. Instead, the use of soft materials could provide a visible advancement. In this work, we propose a soft, sensorized gripper for harvesting applications. The sensing is performed by tracking a set of markers integrated into the soft part of the gripper. Different machine learning-based approaches have been used to map the markers’ position and dimensions into forces in order to perform a close-loop control of the gripper. Results show that force can be measured with an error of 2.6% in a range from 0 to 4 N. The gripper was integrated into a robotic arm having an external vision system used to detect plants and fruits (strawberries in our case scenario). As a proof of concept, we evaluated the performance of the robotic system in a laboratory scenario. Plant and fruit identification reached a positive rate of 98.2% and 92.4%, respectively, while the correct picking of the fruits, by removing it from the stalk without a direct cut, achieved an 82% of successful rate.File | Dimensione | Formato | |
---|---|---|---|
1-s2.0-S0168169923005902-main.pdf
accesso aperto
Licenza:
Creative commons
Dimensione
2.82 MB
Formato
Adobe PDF
|
2.82 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.