Adaptive harvest robots move closer to commercial farming as tomato-picking research improves success rates
New tomato-harvesting research shifts robotic picking from simple fruit detection toward judging the probability of a successful pick, a change that could support wider automation in greenhouse vegetable production.
Research in robotic harvesting is moving beyond simple fruit detection toward decision-making before the robot attempts a pick. Digital Journal, drawing on work from Osaka Metropolitan University, reports that a new tomato-harvesting system does not just identify ripe fruit. It also estimates how easy and safe each pick will be under the actual physical conditions around the plant.
The model, developed under Assistant Professor Takuya Fujinaga, considers fruit position, stem orientation, surrounding leaves and the degree of occlusion before assigning a harvest probability and choosing an approach angle. That matters especially in tomatoes, where fruit grows in dense clusters, ripens unevenly and can be damaged easily if a machine makes the wrong movement.
In testing, the system achieved a success rate of about 81 percent. Some of the successful picks came after the robot changed its strategy mid-task: when a frontal approach failed, it recalculated and attempted a side-angle harvest instead. For commercial horticulture, that is an important shift away from rigidly programmed automation and toward semi-autonomous systems that can adapt to specific crop conditions.
The practical value for growers is not just faster harvesting. A probability-based system can reduce fruit damage, avoid wasting time on low-probability attempts and direct robotic labour toward the most productive tasks. In effect, the technology tries to translate into software the judgement that experienced human pickers already apply when deciding whether a fruit can be removed cleanly.
The article also notes similar movement in Canadian greenhouse production, particularly in Ontario and British Columbia, where tomato growers are investing in automation, robotics and AI-driven monitoring under pressure from labour shortages and higher costs. Related work in machine vision, adaptive gripping and crop analytics is already helping operators decide what should be harvested first. Full human replacement is not presented as imminent, but the combination of people and adaptive robots is increasingly framed as a realistic operating model for intensive vegetable production.