Hi HN,
I’m in the process of developing some new devices for use in agriculture and environmental monitoring.
Before I go much further, I thought I’d see if there are developers who would be interested in using them, and what features/capabilities people might want.
First, a bit of back story as to how I got here. My father started a business in the 1980s, making devices using Motorola 6800 microprocessors and A/D converters to read from “gypsum block” soil moisture sensors [1,2,3]. Though his company did many different things over the years, horticulture and viticulture was always a significant part of the business, and he had a solid product line through the 90s-2000s selling soil moisture readers to wine grape growers and many other kinds of growers, researchers, etc.
I started working with him in 2014 when he wanted to build a version of the device that could send data to a smartphone app and upload it to a web app. My father has now retired, but along the way we picked up several clients and industry partners who love the products we developed. But it’s taken all that time working in the business to really understand all the different use-cases people have, and the kind of device that would tick all the boxes for what people want/need now and what they might want/need in years to come. We had many enquiries from different kinds of users who asked “can you develop a version of your device that does ___”. So I’m trying to come up with a device that caters to all the common (and even many less common) use cases in agriculture and environmental monitoring.
The biggest insights I’ve picked up about this space, and the issues I’d like to address are:
Many products exist that do some kind of environmental monitoring and control, but they are almost always expensive (USD $500+ for a sensor reading/IoT device and another $500 or much more for sensors), are locked down so they can’t be customised, and only have limited support for sensors (these days normally just SDI-12 and pulse counting, maybe other digital inputs like RS-485, but dead ends often arise when a user thinks “now I’d like to read from this other kind of sensor”.
Plenty of hackers/makers/vendors try to build custom solutions using DIY platforms like Raspberry Pi or Arduino, which offer lots of possibilities, but involve a lot of time/effort to get a working product up and running (hardware sensor interface installation, software configuration, software development), and then it turns out to be hard to keep these devices running with adequate power (it’s often impractical to have solar panels and large batteries connected to devices installed in crops, as they can get damaged by machinery, weather, animals, etc.
So, here are the devices and capabilities I have in mind:
==
Device one:
Based on a Nordic nRF5xxx[4], with ARM Cortex-M33 cores and built-in Bluetooth LE, Bluetooth Mesh, Thread, Zigbee
Support for these inputs/outputs/protocols for reading from sensors (some built-in, some via expansion modules):
- RS485/RS422/RS232 (to support many environmental/industrial sensors, irrigation systems)
- SDI-12 (for common environmental sensors e.g, soil probes, weather stations)
- Pulse counting (for tipping bucket rain gauges, irrigation flow meters)
- Dallas one-wire protocol for low-cost temperature sensors
- Analogue impedance inputs for gypsum/tension blocks (many growers and researchers still regard these as the most accurate/useful way to measure soil water availability)
- Analogue voltage inputs for other common environmental sensors including low-cost soil moisture sensors
- Switching outputs for controlling irrigation solenoid valves etc
Optional internet/WAN uplink modules – cellular (LTE-M), Wi-Fi HaLow, Satellite (Swarm or other?), LoRaWAN
Configurable via terminal, smartphone app (Bluetooth LE), MQTT
Scriptable via MicroPython
Ability to run machine learning models using TinyML
Ability to run for extended periods (a year) on batteries (e.g. 18650s) without needing solar power, though solar power and larger batteries can be added where necessary and practical.
==
Device two:
Same networking and sensor capabilities as device one, but with a Raspberry Pi-like compute module and the ability to run Linux (with sensor device/protocol drivers pre-loaded).
==
I’ll be aiming to get the production cost and price as low as possible, so that it’s affordable for anyone to buy these to hack on, including/especially in developing countries.
Regarding machine learning/TinyML support, I’ve wondered if this would really be useful in agriculture, or if it would just be a gimmick that’s ultimately not of any benefit. From speaking to experts on TinyML, it’s considered that it’s only worth doing on an embedded “edge” device like this if it would be too slow/costly/impractical to move data back and forth between the sensor-reading devices and the cloud (or a more powerful machine nearby). But in agriculture there are situations where this would be the case (highly localised plant monitoring to control irrigation). And one particularly interesting suggestion that was made to me was that if you were wanting to roll out a network of many sensors that were low cost/low precision, TinyML could be used to calibrate the date from the many low-precision sensors with a smaller number of higher-cost higher-precision sensors. But another use case could be image analysis, which a few people have told me they would be interested in doing, for things like pollination monitoring and fruit size monitoring. But I also wonder if there are all kinds of other machine learning applications people might come up with if they had a board that supported it.
My intention is to develop something that is powerful and capable and a lot of fun to build on, by taking care of all the boilerplate needs that apply to enviro monitoring (sensor support, network connectivity, power management), and just allow developers to focus on building novel applications.
Please share your thoughts in the comments and feel free to contact me (address in bio) to discuss/enquire further.
[1] https://www.cabidigitallibrary.org/doi/full/10.5555/19410300...
[2] https://acsess.onlinelibrary.wiley.com/doi/10.2136/sssaj1941...
[3] https://journals.lww.com/soilsci/Citation/1942/10000/FIELD_S...
[4] https://www.nordicsemi.com/Products/nRF5340
[1] https://www.lilygo.cc/products/t-sim7600