Robotic model boosts success rate for tomato picking

by Riko Seibo
Tokyo, Japan (SPX) Dec 09, 2025

In many farming regions, a shortage of agricultural labor is driving interest in robots that can harvest fruit crops such as tomatoes. Tomatoes are difficult for machines to handle because the plants produce clusters of fruit at different stages of ripeness, so any harvester must select only ripe fruit while leaving unripe tomatoes on the vine.

Assistant Professor Takuya Fujinaga of Osaka Metropolitan University has developed a method that lets a robot assess how easily it can harvest each tomato before it attempts a pick. The approach is designed to improve both the efficiency and reliability of robotic harvesting in commercial greenhouses and fields.

Traditional research on tomato harvesting robots has focused on detecting fruit and recognizing whether it is ripe. Fujinaga instead framed the task as an estimation problem in which the robot evaluates the ease of harvesting each fruit and then decides how to approach it.

The model combines image recognition with statistical analysis to select the best approach direction for a target tomato. The system processes images of the fruit, the stems, and any parts of the plant that might hide the tomato from view, then uses this information to guide the robot arm.

Key factors include whether a tomato is part of a tight cluster, how the stems are arranged, the position of background leaves, and whether the fruit is occluded by other plant parts. These details influence the robot's control decisions and help it plan a path that is more likely to result in a successful pick.

Fujinaga describes this shift as a move from a detection and recognition model to what he calls a harvest-ease estimation framework. "This moves beyond simply asking 'can a robot pick a tomato?' to thinking about 'how likely is a successful pick?', which is more meaningful for real-world farming," he explained.

To validate the concept, the research team ran harvesting experiments and then used the results to build and test the statistical model. The analysis was based on the outcomes of individual picking attempts and the conditions under which each attempt was made.

In trials, the new model achieved a success rate of 81 percent, which exceeded earlier predictions. About one quarter of the successful picks involved tomatoes that the robot harvested from the right or left side after previous failures when it tried to harvest from the front, indicating that the system can adapt its approach when it encounters difficulty.

These findings show that approaching the same fruit from different directions can significantly change the chances of success. The results also suggest that future harvesters could use real-time feedback to switch strategies and improve performance during operation.

The study underscores how complex fruit picking is for robots, because they must interpret three-dimensional plant structures while working around leaves, stems, and clustered fruit. According to Fujinaga, "This research establishes 'ease of harvesting' as a quantitatively evaluable metric, bringing us one step closer to the realization of agricultural robots that can make informed decisions and act intelligently," he said.

Looking ahead, Fujinaga expects harvest-ease estimation to support systems that can decide whether crops are ready for automated picking. "This is expected to usher in a new form of agriculture where robots and humans collaborate," he explained. "Robots will automatically harvest tomatoes that are easy to pick, while humans will handle the more challenging fruits."

Research Report:Realizing an Intelligent Agricultural Robot: An Analysis of the Ease of Tomato Harvesting