The scan of the east maize block is reviewed. Export the weed targets for the robot.
Robotics
From a map of what's in the field to a list of targets a machine can act on. Our first robotics work turns weed detections into grab and spray points a robot can navigate to.
Weed targets for robots
On a field mission, a segmentation model separates weeds from the crop. An agronomist or operator reviews the map and dismisses false positives. What remains is exported as one target per weed, with its outline and a grab point at the stem or centre.
The detector starts from a pretrained corn model; accuracy improves with field-specific training.
1,284 weed targets ready for the robot.
- Targets
- 1,284, one per weed
- Frame
- UTM 29N, metres
- Format
- ROS weed-target message
| Target | Easting (m) | Northing (m) | Area (cm²) |
|---|---|---|---|
| w-0001 | 512 304.2 | 3 712 845.6 | 38 |
| w-0002 | 512 304.9 | 3 712 846.1 | 21 |
| w-0003 | 512 306.4 | 3 712 844.8 | 54 |
+1,281 more targets in the file
mission_weeds · robot-export12 Oct, 10:05The export is ready. The detector starts from a pretrained maize model, so check a few targets on the row before the first full pass.
- Targets
- One per weed, with outline
- Coordinates
- WGS84 plus local UTM metres
- Formats
- GeoJSON, CSV, shapefile, ROS
- Status
- In development
Export formats
- GeoJSON
- Each weed is a polygon with its grab point, accuracy and local UTM coordinates.
- CSV
- One waypoint per row, a simple format a sprayer or robot controller can read directly.
- Shapefile
- A point shapefile of grab targets, zipped with its projection.
- ROS
- A weed-target message in a UTM frame. The robot navigates in metres and acts at each point.
Status
Robotics is early. The export works over reviewed field missions today; we are looking for equipment makers, research groups and growers to test it on real machines and rows. Spot-spray and herbicide zone maps come from the same detections.