Hello readers!

This week, we’re looking at the affordable, solar-powered robot that is promising to help farmers even off the grid, SGP.32, IoT MVNOs, and more!

A research team from New Mansoura University recently built a solar-powered autonomous robot that rolls through crop rows, photographs leaves, and uses an onboard deep learning model to classify plant diseases in real time. The system — built on a Raspberry Pi, a ResNet-based CNN, and IoT soil and humidity sensors — achieved 97.13% testing accuracy in a tomato field trial. Human expert inspection came in at 95.8%. The robot caught early-stage disease indicators that the agronomists missed entirely. Plant diseases account for up to 30% of annual crop losses globally, and most of the existing detection toolkit still relies on people walking fields and making judgment calls.

The researchers were building explicitly for farmers who can't afford hyperspectral imaging or drone fleets. The full system runs roughly $650 USD per unit at prototype cost, down to an estimated $450–500 at scale. It operates off a 50W photovoltaic panel with battery backup, achieving 86% energy efficiency through a six-hour field session under partly overcast skies. The onboard CNN — compressed from 98MB to 23MB via quantization — processes an image every 220 milliseconds without a cloud connection. That last detail matters because rural IoT deployments can't rely on consistent connectivity, and most agricultural IoT architectures have been designed as if they can.

The IoT-in-agriculture space has spent years building toward this kind of integration, but the bottleneck has almost always been the diagnostic layer: how do you translate sensor data and imagery into a decision a farmer can act on quickly? Smart farming follows an observe-diagnose-decide-act cycle, and wireless sensor networks have handled data collection reasonably well. The harder problem is processing at the point of collection, where bandwidth is thin and cloud round-trips introduce lag. The team's approach — running inference locally on the Raspberry Pi, transmitting only processed results upstream via MQTT — is edge computing applied to crop health, with the added wrinkle that the edge node is mobile and self-powering.

The CNN was trained primarily on PlantVillage, a benchmark dataset captured largely in controlled conditions. The researchers supplemented it with augmented variants to simulate field variability, and the test results support reasonable generalization — but the 2-acre tomato trial is a proof of concept, not a production deployment. A coordinated network of robots covering a large farm, syncing to a shared cloud dashboard, is the roadmap the paper gestures toward. Calibration drift, hardware failures, and keeping AI models updated across a distributed fleet are operational problems that don't show up in a single-session field test.

Cheaper edge hardware, smaller CNN architectures, and lower-power sensors are steadily closing the gap between what precision agriculture can do in theory and what it can do in a field with intermittent connectivity and a tight budget. The most useful takeaway from this paper isn't the accuracy number — it's the architecture: edge inference with local fallback, sensor fusion for contextual diagnosis, and cloud integration for alerts rather than real-time processing. That pattern applies well beyond disease detection, and it's one the industry is still working to standardize.

After slow adoption of SGP.02 and SGP.22, SGP.32 is gaining significant traction in security systems, automotive, body cameras, fixed wireless, and industrial IoT. The key reason is that SGP.32 shifts control of identity.

Read more

Standard H.264 video chews through 4–6 Mbps per camera, driving up cellular data costs and limiting how broadly surveillance can scale.

Semtech's integrated Video Compression Solution — combining Digital Barriers' EdgeVis™ AI compression, multi-carrier Smart Connectivity, rugged AirLink® XR60 routers, and AirVantage® management — reduces cellular video data by up to 90% and delivers TCO savings of up to $2,500 per camera over 36 months.

The best person for your next role might not live near your office — or even in the same country.

And more companies are realizing they don’t need to open entities everywhere just to hire great talent globally.

Instead, teams are using EOR to hire internationally faster, stay compliant, and avoid the operational headache of setting up local infrastructure before they’re ready.

That shift is changing how companies think about growth, hiring, and expansion altogether.

Oyster’s EOR helps companies hire, pay, and support employees in 180+ countries while Oyster handles payroll, compliance, taxes, and local employment requirements.

Talk to experts today

In this episode of the IoT For All Podcast, Matthias Wagner, Founder and CEO of Flux, joins Ryan Chacon to discuss AI-assisted hardware design for IoT. The conversation covers the historical challenges of hardware design, the current capabilities of AI tools, compressing the hardware iteration cycle, integration challenges, the limitations of AI, and enabling IoT innovation.

YouTube • Spotify • Apple • YT Music • Amazon

Interested in becoming an IoT For All Partner? Reach out here!