GOAT INDUSTRIES
A 3D cultivator is capable of manipulating the soil in a bed of seedlings in not only the X axis (forwards and backwards) but also the Y axis (side to side) and Z axis. Conventional cultivators only operate in the X and Z axis.
The next stage is to fit cameras to track the crops and install a hydraulic mechanism to momentarily stop the cultivator in the X axis to enable Y axis movement of the set of claws. The tractor itself will trundle along unperturbed. Watch this space for updates during 2026 !!
The video below shows how we've combined traditional, CPU intensive, Open CV type computer vision systems with modern YOLO26n Ai image inference to try and recognise, and predict, the position of grids of seedlings. The idea is to try and predict when there is a gap between the crops to momentarily stop the cultivator implement in the Y axis and quickly send a set of hydraulically driven rotating claws across the X axis, and back again. The red horizontal X axis line acts as a trigger point and when the system detects that a weighted combination of the inferred Ai driven grid center and the traditional CV grid center overlap, or are close enough to one another, a data point is added to the data array. The system is able to detect the next grid center which is useful if the system misses a trigger point. It's not perfect yet and you can see it does actually miss one prediction, but as long as it gets most of them, this should be ok.
This coming season (2026) we're aiming to get the cultivator actually using the camera system to perform 3D cultivation. We've been busy over the Winter moving the hydraulic control system over to the same enclosure as the camera system and developing a robust communication using USB to communicate between the Nvidia Jetson Naon and the 2 STM MCUs, with the Nano as the master.
Also, we spent a bit of time re-training the YOLO26n model with an additional class - slugs! 7808 images of slugs to be exact. It's going to be fun driving the machine in the dark with some artificial light to detect these creatures. We dont have a big problem with slugs here, but if we wanted to control the population the machine would give them a quick spray of nematode worms upon detection .... no .... not lasers !!
Máté Tóth
John Opsahl
mbasecnc