For OEMs
AI camera system for weeder manufacturers: the AROW box
Camera, compute, AI and blade control in one unit. We agree the details with your engineers, your team builds it into your machine starting from a test box, and our platform keeps the AROW box in every machine you sell updated, maintained and supported remotely.
From first call to series production
You tell us about your machine
Describe the machine, your customers' crops and markets, and what you need from us: cameras only, the full control chain or a complete in-row module. The first conversation commits you to nothing.
We agree the set-up together
Rows, row spacing, actuators, power, interfaces, mounting: our engineers and yours define what AROW delivers and what your machine keeps. You get a sizing, an interface proposal and a realistic timeline.
You build on a test box
A complete AROW unit for your workshop and your field. Your team builds the machine around it (frame, blade units, hydraulics, controls), and we support the build and validate the first prototype with you in a growing crop.
Your machine goes into series production
AROW units for production, training for your dealers and service staff, and the platform in the background: updates, monitoring and remote support for the AROW box in every machine you sell.
Typical project: [weeks – to be confirmed] from first call to field test. The mechanical adaptation usually takes longest, because the AROW side is ready to run.
AROW
What's in the box
The full AROW package for one machine. Figures marked with a dotted underline come from published tests and brochures and are being confirmed by Ullmanna.
- Camera + illumination
One AROW box per row, with one 2D camera and two 3D cameras inside (the two small cameras), plus four high-power flash LEDs. AROW looks for the root point rather than the leaf, so wind, overlapping leaves and low sun do not confuse it. It works by day, at night and in dust.
- Compute and power hardware
A compute module per row and one central Power Box with GPS and connectivity. The models run on the machine; no internet is needed to weed. Power comes from ISOBUS or the tractor battery.
- AI models
Crop-specific detection models, trained on 194 plants and 6.31 billion field decisions since 2021, 2.16 billion of them in the first five months of 2026. A new crop is trained from field images within hours.
- Machine interface and control
Blade control electronics and the signals your machine needs, for pneumatic, hydraulic or electric actuators. AROW drives the in-row blades; your machine keeps its depth guidance, lift and steering.
- Software & updates
Operator tablet with crop selection, status and diagnostics. New models and software reach every AROW box over the air, so your customers never have to take a machine to the workshop for an update.
- Remote support
We see what the cameras see, so we can diagnose faults in the box, fine-tune its settings and help your service team without a field visit.
Specification
- Cameras
- One 2D camera and two 3D cameras (the two small cameras) per AROW box · one box per row
- Illumination
- Four high-power flash LEDs per row · day and night
- Compute
- One AI compute module per row · central Power Box with GPS and connectivity
- Operator interface
- Android tablet via WLAN · crop selection, status, diagnostics
- Power
- ISOBUS or tractor battery
- Rows
- 1 to 16 rows per machine
- Row spacing
- From 23 cm
- Working speed
- Up to 5 km/h in-row in sugar beet · up to 8 km/h in pumpkin
- Decision rate
- About 3 blade decisions per second per row · ~15 ms from decision to blade
- Crop stage
- From the two-leaf stage
- Actuators
- Pneumatic reference design · hydraulic and electric supported
Behind every AROW box
We keep every AROW box updated and monitored
You get more than a box: a complete solution for the box and the infrastructure behind it. Every AROW unit in the field is connected to a platform that Ullmanna operates, so maintenance, updates, monitoring, remote support and data exchange are in place before your first machine ships. You and your dealers look after the machines in the field.
- Maintenance
Every camera, light and compute module reports its condition continuously, so wear and faults show up before the operator notices them.
- Updates
We roll out crop models and software over the air to every AROW box in the fleet, or only to the boxes on one machine when a customer needs a new crop.
- Unit monitoring
Working hours, crops, speed and detection quality for each unit. Your service team and ours see the same data.
- Data exchange
Field images improve the models; agronomic data stays with the operator. Interfaces for your own portal or telematics are available.
Technology
Root point, not leaf
Most vision systems look for leaves. AROW locates the point where the plant enters the soil. That is why wind, overlapping leaves and weeds growing under the crop do not confuse it, and why weeding can start at the two-leaf stage, when hand crews are hardest to find.
- Root-point detection
Every crop plant gets one coordinate. The blades open before it and close behind it, up to three times per second in each row.
- 3D depth
The two 3D cameras read plant and soil height per row, so working depth follows the ground.
- Decisions on the machine
Models run on the row module. No connection is needed to weed; the connection is used for updates, monitoring and remote support.
- Fleet learning
2.16 billion plant decisions in five months of 2026 feed the next models, so every machine benefits from every field.
Root-point detection in six of the 194 trained plants, in field images as the system sees them.
The fleet at peak season
Figures from the first five months of 2026, taken from machines working on farms. The season is still running.
Season 2026 · fleet data from machines at work on farms. All 2026 figures cover only the first five months of the year; the season is still running. Field data 2026 in detail
Autonomous hours per month
3,788 hours in June alone: 37 % of all hours so far, logged in 30 days across 199 machines at once. That is the peak load your customers' machines have to handle.
Top 10 crops by field hours
Ten crops account for most of the hours, and 112 more follow. The models know the main crops in depth and a long list of less common ones.
194
194 plants in one system
The ten crops with the most field hours in the first five months of 2026, plus more than a hundred other crops from arnica and chamomile to tobacco. A crop not on the list is trained from field images within hours.
- Pumpkin
- Lettuce
- Sugar beet
- Tomato
- Celery
- Cabbage
- Broccoli
- Onion
- Watermelon
- Fennel
Show all 194 plantsShow fewer
- Agrimony
- Aloe vera
- Angelica
- Anise
- Arnica
- Artemisia
- Artichoke
- Arugula
- Aubergine
- Baby leaf
- Basil
- Bean
- Bee balm
- Beech
- Beetroot
- Bird's-foot trefoil
- Black medick
- Black radish
- Blackcurrant
- Borage
- Boxwood
- Broccoli
- Brussels sprouts
- Cabbage
- Calendula
- Cape gooseberry
- Caraway
- Carrot
- Cat's-ear
- Catalogna chicory
- Cauliflower
- Cavolo nero
- Celeriac
- Celery
- Chamomile
- Chard
- Chervil
- Chia
- Chickpea
- Chicory
- Chili pepper
- Chinese cabbage
- Chives
- Cleavers
- Common daisy
- Common mallow
- Coriander
- Corncockle
- Cotton
- Courgette (Ambassador)
- Cowslip
- Cress
- Curry plant
- Dahlia
- Daikon radish
- Dandelion
- Dianthus
- Dill
- Douglas fir
- Dropwort
- Endive
- European goldenrod
- Evening primrose
- False brome
- Fennel
- Field wood-rush
- Fir
- Flat-leaf parsley
- Garden pea
- Garlic
- Grapevine
- Great burnet
- Greater burdock
- Greater burnet-saxifrage
- Greenhouse spinach
- Gumweed
- Hemp
- Hogweed
- Hop trefoil
- Horseradish
- Hound's-tongue
- Kale
- Kalette
- Kohlrabi
- Large thyme
- Lavender
- Leek
- Lemon balm
- Lemongrass
- Lesser stitchwort
- Lettuce
- Liquorice
- Lovage
- Maize
- Marigold (Tagetes)
- Marjoram
- Marsh mallow
- Meadow bellflower
- Meadow salsify
- Meadowsweet
- Melilot
- Mouse-ear hawkweed
- Mugwort
- Mullein
- Mustard
- Nasturtium (Alaska)
- Nasturtium (Tom Thumb)
- Onion
- Oregano
- Oxeye daisy
- Pak choi
- Parsley
- Parsnip
- Passion flower
- Pea
- Peach-leaved bellflower
- Peanut
- Peony
- Pepper
- Peppermint
- Pointed cabbage
- Potato
- Prickly poppy
- Primula
- Proso millet
- Pumpkin
- Purple coneflower
- Quinoa
- Radicchio
- Radish
- Ragged-robin
- Rapeseed
- Red campion
- Red clover
- Rhodiola
- Rhubarb
- Ribwort plantain
- Rice
- Romanesco
- Root parsley
- Rosemary
- Rough chervil
- Rough hawk's-beard
- Rough hawkbit
- Runner bean
- Sage
- Savory
- Savoy cabbage
- Scotch thistle
- Shallot
- Small bugloss
- Smooth hawk's-beard
- Sorghum
- Sorrel
- Soybean
- Spinach
- Spring onion
- Spruce
- Squash
- St John's wort
- Stevia
- Stinging nettle
- Strawberry
- Sugar beet
- Sugar melon
- Sunflower
- Swede
- Sweet potato
- Sweet william
- Tansy
- Tarragon
- Thyme
- Toadflax
- Tobacco
- Tomato
- Turnip
- Valerian
- Viola (Sorbet)
- Wasabi
- Watermelon
- Weasel’s snout
- Weld
- Wheat
- White bean
- White campion
- White mullein
- Wild carrot
- Wild rocket
- Wild teasel
- Wood avens
- Wood sage
- Wormwood
- Yarrow
- Zucchini
As of season 2026, exported from the Ullmanna fleet management. Includes crops, herbs and medicinal plants as well as the weeds and wild plants the system tells apart from them.
Crop not on the list? We train it from field images within hours.
Nine manufacturers already build on AROW
From two-row vegetable hoes to twelve-row sugar beet frames, from Central Europe to Japan: nine machinery manufacturers build the AROW box into their own machines. Widths, actuators and crops differ; the vision system inside is the same.
To see a machine with AROW in the field, talk to us. We will connect you with the right manufacturer for your region.
Why manufacturers choose AROW
- AI in-row capability without building a vision system
Seeing each plant and deciding within 15 milliseconds is the hard part, and it is already done. You add it to a frame you already sell or to a machine you are planning.
- Field-proven for four seasons
199 machines logged 10,250 hours in the first five months of 2026 and 31,000 hours since 2021, from Austria to Canada and Japan. This is a working fleet, well past the pilot stage.
- A crop dataset that grows every season
6.31 billion plant decisions since 2021, 2.16 billion of them in five months of 2026, feed the models your machine runs on. Each field improves the models for every machine.
- New crops in hours
194 trained plants today. When one of your customers grows a niche crop, we train it from field images, with no development work on your side.
- Day, night and difficult fields
With its own flash lighting and root-point detection, AROW works at night and under high weed pressure, even in "bad" fields.
- From one row to sixteen
The same package scales with the frame. Camera, compute and power per row, one Power Box per machine.
- Over-the-air updates
Models and software improve across the whole fleet without a workshop visit.
- Our engineers work with yours
Mounting, control interface and field validation, on your machine, from the first call to series production.
- Experience from our own fields
Built by a family of farmers and engineers who have run the machines themselves since 2021.
Questions manufacturers ask
01How does a project with Ullmanna start?
You contact us and describe your use case: the machine, the crops and markets of your customers, and what you need from us. We then agree the set-up together and you receive a test box to build your solution around.
02How many rows can AROW control?
From 1 to 16 rows on one machine. Every row gets its own camera unit and compute module; one Power Box supplies the whole frame.
03Which crops are supported?
194 trained plants as of season 2026, from sugar beet, pumpkin and lettuce to tobacco, arnica and chamomile, plus the weeds and wild plants the system tells apart from them. A crop that is not on the list is trained from field images within hours.
04What does AROW connect to on our machine?
AROW controls the in-row blades via its own control electronics and takes power from ISOBUS or the tractor battery. Depth guidance, lift and inter-row tools stay with your machine. The interface is defined together with your engineers.
05Which actuators does it drive?
Pneumatic blade units are the reference design; hydraulic and electric actuation are supported. Response time from decision to blade is about 15 ms.
06Who looks after the machines once they are with customers?
You and your dealers look after the machines in the field and remain the contact for the farmers who use them. Ullmanna is responsible for the AROW box and the infrastructure behind it: maintenance and health monitoring of every unit, over-the-air updates for models and software, monitoring of working hours and detection quality, data exchange with your own systems if you want it, and remote support for the box, which also helps your service team.
07Who owns the field data?
The operator owns the agronomic data of their field. Anonymised plant images improve the models that every AROW machine runs on. That is why a crop trained in Austria also works in Canada.
08How long from first call to a working prototype?
Typical project: [weeks – to be confirmed] from first call to field test. The mechanical adaptation usually takes longest, because the AROW side is ready to run.
Contact
Talk to us about your machine
Describe your use case: machine, rows, crops, markets and what you need from us. You get a sizing, an interface proposal and a realistic timeline, and we discuss the test box.