Teaching a robot to harvest tomatoes

Published on
June 9, 2020

​Harvest a pepper or a tomato. Or remove a leaf. Simple actions you would say. But programming a robot so that he (or she) can do it is almost impossible. That is why the Greenhouse Horticulture Business Unit of Wageningen University & Research is investigating how a self-learning robot can be of help in a greenhouse.

Robots are starting to be available in the greenhouse. They can work 24/7 in high humidity and temperature conditions, operating in very hygienic conditions. Although, they can do repetitive jobs with high accuracy and time performance, in the greenhouse their use is limited to crop monitoring or to a very specific single task with a well-defined variety of crops, e.g. strawberries harvesting. But a different task requires a different robot. A robot that can learn from a human is more flexible.  

That is why Wageningen University & Research is investigating possible applications of learning robots in agri-food, such as greenhouse, open fields, livestock, food processing or aquatic farming, thanks to the Knowledge Base Robotics project of the program Data-Driven High Tech (DDHT).

How the robot learns

The research into the possibilities in horticulture is done at the Greenhouse Horticulture Business Unit. The researchers show the robot small actions, over and over again, each time performed slightly differently or from a different angle. The robot learns how to perform a task, training itself for weeks using reinforcement learning approach.

In this way the robot learns tasks that are natural for humans but very difficult to program, as to remove a leaf when a ripe tomato is hidden behind it or to avoid contact with a human in its path. Furthermore, WUR is investigating how robots learn to respond to difficult conditions as the limited workspaces in the greenhouse due to the presence of the crops and the constantly changing lighting conditions.

This project is funded by the Ministry of Agriculture, Nature and Food Quality.​