In the high-stakes world of modern greenhouse agriculture, trade-offs between speed, crop safety and biosecurity often determine whether automation succeeds or fails.
Rather than forcing a path through heavy foliage or risking fruit damage, Eternal.ag has taken a deliberate, integrity-first approach to robotic harvesting: cut the stem, never touch the fruit and skip occluded trusses until a 100% safe path is guaranteed.
In a conversation with The Packer, Roel Janssen, chief business development officer of Eternal.ag, explains how its modular, simulation-trained harvesting robot tackles complex canopy occlusions, mitigates viral cross-contamination and prepares for its imminent commercial footprint in North America.
The Packer: How does your artificial intelligence vision handle severe occlusion (where 50% or more of a ripe tomato cluster is physically blocked by leaves or stems)? Does the robot have a way to gently brush aside foliage to get to the fruit or, does it simply skip obstructed clusters?
Janssen: Occlusion is one of the most challenging technical problems in autonomous harvesting and one that the industry as a whole continues to work on for various crops. Our current focus is on delivering reliable, high-quality harvesting in scenarios where the robot can make confident decisions without compromising the crop or the plant. However, it’s important to distinguish between fruit occlusion and peduncle occlusion.
The strong uniformity of greenhouse crop growth means that the bottommost tomato truss reaches harvest ripeness in line with the regular weekly harvest schedule, so the robot does not need to assess the color of individual fruit. As a result, the real challenge of occlusion isn’t identifying which truss to harvest; it’s reaching the peduncle and making a clean cut.
Our robot harvests one truss at a time. It uses a stereo camera to build a 3D model of each tomato truss and its surrounding canopy before selecting a harvesting approach. Before every cut, the software determines whether there is enough clear space around the peduncle for the gripper to reach it without contacting any fruit. If a safe, collision-free path doesn’t exist — typically because leaves or stems are heavily obstructing access — the robot deliberately skips that truss rather than risk bruising fruit or damaging the plant. That’s a design decision in favor of crop and plant integrity, not something we’ve overlooked.
In current deployments, this puts us at up to 85% autonomous harvest coverage. The remaining share, the trusses obstructed by occlusions, is currently harvested manually and is the focus of our active development. We are already actively developing a solution to also harvest the remaining 15% and expect to have that market-ready next year. Our goal is clear: a fully autonomous greenhouse with 100% harvest coverage and no manual intervention.
Greenhouse biosecurity is incredibly strict due to viruses like tomato brown rugose fruit virus, or ToBRFV. If a robotic end-effector touches an infected plant, it risks transferring the pathogen to every subsequent plant down the row. How does the harvester prevent cross-contamination? Is there an automated sterilization mechanism built into the cycle?
Our sterilization is built into the harvest cycle: The end-effector is sterilized on the robot itself, which is typically done at the end of every line. That keeps the row-to-row transmission risk you’re describing (one infected plant seeding the rest of the row) under control within the operating routine, rather than depending on manual intervention. On the detection side, the robot already carries computer vision as part of how it locates and harvests each truss. Over time, that same vision capability can be extended to scan for signs of disease.
How does the robot’s end-effector adapt its physical grip pressure depending on the variety’s skin thickness? What is your current rate of mechanical crop damage (bruising, skin tears) during harvesting?
Our robot doesn’t grip the tomato. It grips the peduncle — the stalk carrying the truss — and cuts there, then places the whole truss in a basket. Grip force is set for reliably holding and cutting the peduncle, not calibrated against the fruit. Because the gripper never contacts the fruit, variety skin thickness isn’t a factor in our harvesting.
The North American market is listed as an imminent commercial objective. How soon can U.S. greenhouse growers expect to see Eternal.ag’s harvester piloting or commercially operating on American soil?
Working in North America isn’t a future ambition for us; it’s already happening. We’re on the ground in both the U.S. and Canada, where we’re in advanced discussions with growers. Our plan is to have our first commercial operation running early next year.
U.S. greenhouses often operate at different scales and climates from European equivalents. Will the harvester require physical modifications to operate in U.S. facilities, or does your simulation-first training allow it to adapt instantly?
Our robots are designed for global operation from the outset, so U.S. facilities don’t require physical redesign to get started, except for the difference in power source voltage between the EU [European Union] and U.S. Each deployment involves some on-site work (tuning the system to each specific operation and handling differences in scale or climate); this is handled in the initial commissioning phase rather than through physical modification of hardware.
Two things make that possible. First, the practical reality of the industry: Modern high-tech greenhouses are built to broadly comparable standards worldwide; row spacing, gutter heights, growing systems and rail infrastructure tend to follow the same proven conventions, whether the greenhouse is in the Netherlands, Germany or the U.S. A robot that works in one is operating in a familiar environment in the others. Second, our simulation-first approach means the system is trained across a wide range of conditions before it ever enters a specific facility, which helps it adapt to local variation in crop layout and canopy rather than needing to be rebuilt for each site.
Tomatoes are your first commercial target. Are you already testing other high-wire greenhouse crops (such as cucumbers or peppers) with your modular system, and are those also part of the U.S. rollout plan?
Tomatoes are our first commercial crop, and that’s where our focus is today. Cucumbers are next on the roadmap, and in principle, all high-wire crops are within reach of the system as the same architecture extends across them. Looking a little further ahead, strawberries are also on our radar. For now, though, the U.S. rollout is built around tomatoes. Additional crops will follow as each is proven at scale, rather than launching everything at once.
