AI
Precision Farming Needs Edge AI
About 25% of global crop production is lost to post-harvest waste. In the Netherlands, the average farm produces 10 million fruits per year, meaning roughly 2.5 million apples or pears never reach the market at competitive prices. Aurea Imaging targets that loss with precision thinning: a tractor-mounted camera system detects blossoms in real time, builds a per-tree flower density map, and drives an automated sprayer that applies a saline solution only where needed, reducing fruit count while increasing size and uniformity.
TreeScout: What Runs on the Tractor
The TreeScout is a ruggedized Nvidia Jetson board with camera sensors and a tractor GPS for geolocating each tree. It processes 1080p video at 15 frames per second, generating about 25 megabytes of raw image data per second. Tractors have no Wi-Fi, so the device relies on 4G and LTE. That bandwidth constraint forces all ML inference onto the Jetson. Only lightweight blossom map data travels to the cloud. The device also connects to a universal terminal inside the tractor cabin so the grower can operate it without engineering help.
From a Single Python Script to K3s
Three years ago, the TreeScout ran one Python application inside a Docker container with a watchdog process. The tractor connects directly to the battery, so hard power cuts are routine. The watchdog failed along with everything else, leaving no visibility into application state. The team refactored into microservices on a single-node K3s cluster running on the Jetson. A proof of concept shipped in three to four months. The current setup is now in its third blossom season, with clear root-cause analysis and rapid bug-fix deployment.
Kairos: Immutable OS for Far-Edge Devices
Kairos, a CNCF sandbox project, provides the host operating system across the TreeScout fleet. Its read-only root filesystem blocks unintended changes: an apt install attempt simply fails. Trusted boot combines secure boot with systemd measurement of the kernel, initramfs, and kernel command line. If anything changes, the device refuses to boot rather than expose potentially sensitive data. The Kairos operator lets engineers trigger full OS upgrades from Kubernetes, with canary rollout controls to limit how many nodes update at one time.
Over-the-Air Upgrades Across 60 Devices
The fleet grew from 25 devices last year to 60 this year, with some now deployed in the US. Keeping Nvidia JetPack, CUDA, camera firmware, and modem drivers current requires frequent host-level changes, none of which live in application containers. Kairos handles those updates over the air from the office. No one needs to travel with a laptop and a cable to flash hardware. Provisioning now means sending an OS image to the board supplier for flashing, then supplying only device-specific certificates, which shortened time-to-customer-ready significantly.
Open Source Feedback Loop: Kairos and Aurea Imaging
Real-world use on Nvidia Orin NX hardware let the Kairos team move Jetson-specific setup logic into a shared component called kairos-init. Aurea Imaging’s Dockerfile shrank from roughly 150 lines to around 30. Any new user targeting Orin NX now inherits those improvements. The relationship also pushed the Kairos project to formalize its governance model, adding a voting mechanism for new contributors after William Resour joined as a maintainer. Both sides confirmed where the project works well and where it needs changes, feedback the team rarely gets when things are running smoothly.
Notable Quotes
all these wouldn’t have been possible without us trying daring technologies like K3s, Kyros uh and uh reaching out to the open source community. Jordan Karapanagiotis · ▶ 22:20
tractors don’t come with Wi-Fi on board yet. So we use 4G and LTE networks. Jordan Karapanagiotis · ▶ 04:47
if you try to do an apt install it will simply tell you I’m sorry but I can’t because I cannot write on the disk. Mauro Morales · ▶ 12:23
There’s nothing sadder than having a cool solution but nobody adopts it, right? Mauro Morales · ▶ 22:46
Key Takeaways
- A single-node K3s cluster on a Jetson replaced a fragile Python watchdog in three to four months.
- Kairos immutable OS blocks unauthorized changes and enables over-the-air upgrades without field engineers.
- The fleet doubled from 25 to 60 devices in one year, targeting a 20% reduction in crop waste.
About the Speaker(s)
Mauro Morales is a Staff Engineer at Spectro Cloud and a maintainer of the Kairos project, a CNCF sandbox immutable Linux distribution for edge Kubernetes. He also organizes the Cloud Native Brussels community group.
Jordan Karapanagiotis is a software engineer at Aurea Imaging focused on IoT systems and remote sensing. He has spent six years in AgriTech startups and currently leads edge software development for the TreeScout device, including the move to containerized, Kubernetes-managed deployments.