I’d love to ask the community a simple question.
If you were starting an open-source agricultural project today with a limited budget, what would you build first?
I’m interested in seeing where the community believes open-source agriculture can have the biggest impact over the next five years.
@richardseshie68
Instead of starting a new open-source agricultural project, I would suggest making simple upgrades to our existing FarmBot Open Source Technology. These would be my suggested improvements:
Modernize FarmBot Weed Detection Software
For a $200k software upgrade, Modern AI methods could dramatically outperform FarmBot’s current Weed Detection Software because they rely on state‑of‑the‑art deep learning architectures rather than the computer‑vision heuristics FarmBot uses today. The existing system depends on manually tuned image filters, color thresholds, and blob detection to identify weeds — an approach that struggles with variable lighting, soil backgrounds, plant overlap, and seasonal changes. Modern detectors such as YOLOv11n, YOLOv11s, NanoDet, and SSD‑Lite achieve 94–98% precision and recall in real agricultural environments, even on low‑power devices, thanks to their ability to learn robust visual features directly from thousands of annotated images. These models also run efficiently on embedded hardware like Raspberry Pi 5 or Jetson Orin, enabling real‑time inference in the field.
A modern FarmBot AI pipeline would be built around three components: (1) a lightweight object‑detection model, such as NanoDet or YOLOv11n, trained on a diverse dataset of weeds and crops; (2) an edge‑optimized inference engine (TensorRT, ONNX, or OpenVINO) running on a Jetson or Coral TPU; and (3) an autonomous decision layer that triggers mechanical or laser weed removal without human review. NanoDet can achieve 98.6% precision and 97.7% mAP, while YOLOv11n reaches 207 FPS on workstation hardware and 3–5 FPS on Raspberry Pi 5 — fast enough for FarmBot’s slow, deliberate movements. These architectures also outperform transformer‑based detectors like RT‑DETR and Faster R‑CNN, which are too slow for real‑time robotics, often exceeding 3 seconds per frame.
Other agricultural robotics systems already demonstrate how powerful these modern AI methods can be. Commercial in‑row weeding tractors widely use YOLO‑family detectors for real‑time crop/weed discrimination at high speeds, with YOLOv8 and YOLOv9 running 110–350 FPS on Jetson hardware. These systems show how deep learning enables precise actuation within milliseconds, even when plants overlap or lighting changes. YouTube examples include Blue River Technology’s See & Spray system, which uses deep learning to identify weeds and selectively spray herbicide, and Carbon Robotics’ LaserWeeder, which uses high‑speed computer vision to target weeds with lasers. Rolling out a modern AI upgrade for FarmBot first in the North American market makes sense due to existing user density and hardware availability, followed by global deployment to gardens and small farms everywhere once the models are retrained for regional weed species and environmental conditions.
Carbon Robotics: https://youtu.be/X3ftki5CpDM?si=GpKDKnVRka1FSRiB
Blue River See and Spray: https://youtu.be/XH-EFtTa6IU?si=Pu4afTNGorfrTNcJ
FarmBot Untethered and Solar Powered
With roughly $200k, FarmBot’s hardware and software could be redesigned into a more, field‑capable system that operates off‑grid and cultivates far larger areas than the current bed‑mounted gantry. The solar upgrade is straightforward: FarmBot’s existing power draw is low enough that a compact 300–1000W solar array and a LiFePO₄ battery pack could provide continuous operation. The only remaining engineering puzzle is water delivery, which could be solved through onboard tanks, drip‑line docking stations, or lightweight hose‑following systems (The hose system which would make the system revert back to a type of tethered configuration). Although water is an important commodity, in the mid and Northern USA it is not such a precious commodity like it is in California, and so less targeted water drip lines could be used.
LASER Weeding Refined
A second major leap comes from integrating a laser‑based weed‑removal tool into the Universal Tool Mount. With an additional $150k, the prototype laser design could be refined into a safe, reliable, fire‑resistant module. Modern agricultural laser systems already demonstrate how to eliminate weeds without chemicals, blades, or mechanical disturbance. Carbon Robotics’ LaserWeeder is a strong example—its YouTube demonstrations show high‑precision weed destruction using computer vision and controlled laser pulses. The key to making this safe for small gardens is engineering the beam‑control system: limiting power, adding thermal sensors, using short‑pulse lasers, and integrating automatic shutoff logic. Once refined, a laser tool becomes a permanent, maintenance‑free weed‑control solution that works day or night, in any soil condition, without consumables. There are so many advantages over a mechanical weeding system.
Laser Weeding Prototype by Liberty University: https://youtu.be/HoN3W9uYj6Q?si=sdFNbFLAFX-Xr4rm
Together, these two improvements—solar autonomy and laser weeding—would transform FarmBot from a stationary garden robot into a scalable, open‑source agricultural platform capable of supporting gardens and micro‑farms across North America. After validating the system in this market, the technology could be rolled out globally, with regional AI models trained for local weed species and environmental conditions. This approach aligns with the broader trend in open‑source agriculture: robotics and AI are the highest‑leverage investments over the next five years, enabling small autonomous units to produce more food with less labor, less energy, and fewer inputs.