A small machine,
working row by row.
An autonomous ground robot that sows seeds, spot-sprays weeds, places fertilizer, and removes the unwanted — one plant at a time, all night, without a driver.
Four jobs. One robot.
Same chassis, different end-effector. The brain stays the same.
Precision sowing
Plant seeds at the optimal depth and spacing — adjusted in real-time from on-board moisture and density sensors.
Targeted spray
See-and-spray pesticide and herbicide. 90% less chemical because we hit the leaf, not the field.
Spot fertilize
Place N-P-K and micronutrients exactly where the plant will use them. Cuts run-off, cuts cost.
Mechanical weeding
Physical weed removal — no chemicals at all. For organic farms and specialty crops where spray isn't an option.
How the robot sees.
A YOLO-class vision model on a Jetson, running at 30 fps under the canopy.
Detect, classify, decide — before the wheel passes.
Stereo + RGB cameras feed an on-device model trained on weed atlases (DeepWeeds, CottonWeedDet12) plus farm-specific data we collect during the first week of every pilot. The robot knows the difference between a soybean and a pigweed by week two.
- 30 fps inference at the edge — no cloud round-trip.
- Per-plant treatment record, written back to your farm OS.
- Active-learning loop: misses become next week's training data.
- Fail-safe: if confidence drops, the robot stops, photographs, asks.
Off-the-shelf parts. On-farm fixable.
The first version is built on a commodity 4WD chassis, a $250 Jetson, and a $99 camera. If a wheel breaks in West Texas, you don't ship the robot back to a lab. You replace the wheel.
- No proprietary connectors. XT60, M3 bolts, standard 14AWG.
- OTA software updates, signed and reversible.
- Hardware E-stop within arm's reach on the chassis.
- Bring-your-own implements: seed hopper, sprayer, weeder.
What we're targeting.
Design targets from internal testing and partner farm projections.
We're picking three farms
for the first pilot.
Row crops, orchards, or specialty — if you have land and a problem that's worth a robot's time, we'd like to talk.