Tomato-Picking Robot Arm Aims to Take Over Harvesting Tasks

The developer is currently testing the robot at home and hopes to eventually trial it at the Tolmachevsky greenhouse complex. The project grew out of a TRK hackathon, where students were challenged to create more than a simple robotic arm. The goal was to build a system that uses a camera and a neural network to identify fruit, determine its location and ripeness, and then pass that information to a robot for harvesting. The team also plans to adapt the technology for other vegetable crops. Anton Vlasenko shared details of the project with ITRussia.

– What is the core idea behind your project, and what are its main components?

– It includes much more than the manipulator itself that picks the fruit. The system combines a mobile platform, a computer vision module, sensors, and software. A camera identifies a tomato, the neural network determines its position and ripeness, sensors verify the distance and confirm that the fruit has been successfully gripped. Meanwhile, the software controls the manipulator’s movements and collects harvest data. In practice, this is not just a robotic arm but a complete hardware-and-software platform designed to automate both harvesting and crop monitoring.

– How large is the team behind the project?

– During the hackathon, I developed the core concept and the first version of the solution almost entirely on my own. Once the project began evolving into a startup, we assembled a team. Today, the technical side of the project is largely driven by two people. I handle the software, computer vision, control logic, and the overall system architecture, while Yakov is responsible for 3D printing, assembly, hardware soldering, and further refinement of the physical platform.

In addition, the project includes a sales and deployment specialist, a lawyer, and a 3D modeler. So while the team is relatively small, we are working to cover the entire path from prototype to a product that can ultimately be deployed at commercial agricultural operations.

– What role does AI play in the project?

– The neural network analyzes images from the camera, identifies tomatoes among the leaves and stems, determines their position, and helps the system decide which fruit the manipulator should approach. We also use image analysis to assess ripeness. Part of the image dataset was collected and labeled by our team.

– Your robot does more than simply pick red tomatoes – it also analyzes ripening timelines. How does that work?

– The robot evaluates each tomato using images captured by the camera. It analyzes color, size, and visual indicators of ripeness. The algorithm also takes the tomato variety into account, because different varieties can differ in color, shape, size, and ripening speed. In other words, the system does not simply look for red fruit – it determines the stage of ripeness based on the characteristics of a specific variety. That allows it to decide whether a tomato is ready to be harvested immediately or should be left to ripen further, while also collecting crop-performance data.

– Why did you choose tomatoes and commercial greenhouses in particular? How strong is demand for a solution like this today?

– We chose tomatoes for a reason. First, they are one of the most widely grown crops in commercial greenhouses, and harvesting still depends heavily on manual labor. Second, tomatoes present a meaningful automation challenge: fruits can vary in size, shape, and color, may be partially hidden by leaves, and can grow at different heights. If the system can reliably handle tomatoes, it can later be adapted for other crops as well.

We also deliberately focused on commercial greenhouses because they offer a more predictable environment. Rows are organized, pipe-rail systems are already in place, logistics are straightforward, and lighting and growing conditions are relatively consistent. For a robot, that is a much more manageable setting than an open field.

Demand for solutions like this is likely to be strong. In Russia, the area devoted to winter greenhouses increased from 3,350 hectares in 2024 to 3,460 hectares in 2025, while greenhouse vegetable production reached approximately 1.65 million metric tons in 2025. That means the market is already substantial and continues to expand. Yet, tomato harvesting remains highly dependent on manual labor, and the agricultural sector is already facing workforce shortages. As a result, harvesting automation is no longer just an interesting technology – it has become a practical necessity. Another factor is that Russia still lacks a commercially available greenhouse-harvesting robot that operators can simply purchase and deploy.

– Can the technology be adapted for other crops?

– Right now, our focus is on tomatoes because they are one of the key crops grown in commercial greenhouses, but the underlying architecture is not limited to tomatoes alone.

The robot can eventually be adapted for cucumbers, peppers, eggplants, leafy greens, and other greenhouse-grown crops. According to industry reports, cucumbers and tomatoes account for the largest share of greenhouse production in Russia, although peppers and eggplants are also widely cultivated.

Adapting the system would require retraining the neural network for a new crop, modifying the ripeness-assessment algorithms, and, if necessary, redesigning the gripper. In other words, we would not need to rebuild the entire platform from scratch. Instead, individual modules could be tailored to a specific crop and the operating conditions of a particular greenhouse.

– Have you estimated the robot’s economic payback?

– Based on our calculations, a single robot can match the harvesting output of roughly six workers. During the first three years, operating the robot would cost a greenhouse operator about 330,000 rubles per month (approximately $4,200), including maintenance and the allocated cost of the equipment. After that, ongoing expenses would consist primarily of servicing and electricity, totaling around 150,000 rubles per month (about $1,900). A harvesting crew would cost significantly more.

The robot is not intended to replace the entire greenhouse workforce, because employees also perform crop-care tasks. However, it can substantially reduce the labor required for harvesting, and in the future it may also help automate on-site packaging.

– Are there other companies in Russia developing similar systems? What sets your robotic manipulator apart?

– There are hardly any market-ready commercial tomato-harvesting systems in Russia. One example is OKrobot, but it has primarily been tested in greenhouse cucumber production. There is also AgroRobotics Alpha, a tomato-harvesting robot under development. However, it remains a prototype undergoing trials rather than a market-ready product available for large-scale deployment.

Our advantage is that we are designing the system specifically for Russian commercial greenhouses, taking into account their dimensions, infrastructure, operating conditions, and day-to-day production requirements. In addition to harvesting, we are building a software layer that includes ripeness assessment, harvest statistics, configurable operating parameters, and crop analytics. Another important consideration is phytosanitary safety during harvesting. That matters because a greenhouse robot must do more than pick fruit efficiently – it also needs to minimize the risk of spreading diseases from one plant to another.

– What stage is the project at today?

– We are currently refining the overall system design, including the structure, dimensions, mobility mechanics, and the way the manipulator will operate not as a standalone prototype but as part of a complete mobile platform for greenhouse use. At the same time, we are preparing applications for grant funding, because the next major milestone is to build a more advanced prototype and begin testing it in real greenhouse conditions. Right now, the priority is to bring the project to a stage where it can be tested safely and effectively in a commercial greenhouse environment. The next step is to further develop the platform, integrate it with the manipulator, and then move on to pilot trials at a greenhouse complex.