When Robots Take Over the Apple Harvest

On Lake Constance, the apple harvest still involves a great deal of manual labor. According to the Lake Constance Fruit Growing Competence Center, 246,700 metric tons of apples were harvested in 2024—equivalent to about 1.6 billion individual picking operations. Digitalization and automation can therefore make an important contribution to sustainable agriculture.

The vision of the two universities, RWU and DHBW Ravensburg, is to develop a harvesting robot. At the DHBW’s Friedrichshafen Technology Campus, researchers therefore began by working on a simulation of the system. This simulation allowed them, for example, to test how the sensors functioned under various weather conditions. In addition, various safety scenarios were tested virtually, such as the detection of people. Working with a simulation saves time and costs in the development of the actual robot. 

AI Determines Ripeness and Predicts Harvest Yields

Under the leadership of Professor Dr. Jörg Eberhardt, researchers at RWU developed an artificial intelligence system capable of determining the ripeness of apples. It forecasts harvest yields and determines the ideal harvest time. The information is made available to fruit growers in a user-friendly format via an app. This system was developed and tested by Professor Dr. Markus Lauterbach of RWU and his team. The robot autonomously navigates its routes using cameras and GPS signals. Professor Dr. Thomas Dietmüller and research assistant Pranav Gangavarapu were responsible for the project at DHBW. 

InMach is contributing the mobile system platform, while vision&robotics is providing additional robotics expertise. The Federal Ministry of Economics is investing more than half a million euros in this forward-looking project. The first field test was conducted at the Bernhard fruit farm in Kressbronn. During the demonstration, RWU staff members Fabian Heller and Chaitanya Grandhi, together with Charan Ram Akupati from InMach, initially moved the prototype through the orchard at walking speed. 

A production-ready harvesting assistant could be ready in three years

Ultimately, the system is designed to operate autonomously around the clock. Conditions in an apple orchard are challenging: changing light, wind, and weather, as well as branches, leaves, and people in the work area. Unlike in an orderly logistics warehouse, the robot must reliably navigate a constantly changing environment—even at night. “Our vision is to have the robot operating with pilot customers starting in 2027,” says Jörg Eberhardt. According to InMach’s estimates, a production-ready harvesting robot could be ready in about three years.

This technology is not only useful for harvesting. Applications are also conceivable for winter pruning, optimizing plant protection, or assessing pest infestations. The Bernhard Fruit Farm continues to make its orchard available as a test site. This ensures that the region remains a real-world laboratory for exploring how AI, simulation, and robotics can facilitate work in fruit growing in the future.