The intelligent robots from Wageningen AgroFood Robotics learn – from people – to perform multiple tasks. According to Wageningen University & Research this makes the processing of fruit and vegetables even more (cost) efficient and increases the added value of fresh produce. The first demo application will be ready in the summer of 2021.
Sorting, grading and assembling fresh fruit and vegetables has so far been done mostly manually all over the world. “It would save companies a lot of time and money if robots could take over these time-consuming, boring and self-repeating tasks from employees,” says Aneesh Chauhan, expert leader of the Computer Vision and Robotics group at Wageningen Food & Biobased Research and member of the Agro Food Robotics team. “And it would be even more efficient if one robot could perform multiple tasks.”
Robots that can learn
With this in mind, Wageningen Agro Food Robotics launched the Autonomous Robots for Agrifood Processes project in 2019. One of the goals of this four-year project is to develop intelligent robots that learn to perform multiple tasks, from picking and sorting different fruits and vegetables to separating specimens of the same species based on size and shape, for example.
“Fruit and vegetable processors can use such robots to improve the cost-benefit ratio of their work,” says Chauhan: “For example, consider a robot that processes oranges but can easily switch to processing strawberries, and can immediately sort this fruit for sweetness. The manufacturer can then sell the sweeter strawberries at a premium price. And he can give customers a guarantee that his products have not been touched by human hands, a distinct advantage in the times of a pandemic.”
Recognising human activity
The Wageningen robots are equipped with cameras that capture the movements of a human expert performing picking and sorting tasks. “The recognition of human activity includes data about the person, the environment and the object being handled,” explains Chauhan. “Built-in computer vision and deep learning technology allow the robots to capture this information and learn new tasks by copying them from humans.”
The researchers have now successfully used the robots in multiple situations, including picking and sorting tangerines of different sizes, separating ripe from unripe bananas, sorting intact and damaged cucumbers, and separating different types of fruit on a table.
Demo application
The first demo application will be ready in the summer of 2021, a global first in robotics for agrifood. “An innovation like this requires technological expertise of robotics and machine learning technologies, as well as extensive knowledge of product physiology and food processing. Here in Wageningen, these disciplines work closely together,” says Chauhan.
More advanced sensors
The next step is to test the robots in a (pilot) factory. The researchers also want to make the robots even smarter, by equipping them with more advanced sensors for measuring sweetness, juiciness and firmness. Chauhan, meanwhile, is already thinking a few steps ahead: “If people start to see robots as colleagues, or as students, what would the future look like?”
