Discussion and help for the open source firmware of the RAK2270 Sticker Tracker
- Here you can find help for problems with the open source code firmware
Hi @beegee I am very interested in the RAK2270. I have three on order. I would really like to learn how to control sending the acceleration information.
Later I would like to try some machine learning with the acceleration data. Presently I can install the boards manager on the old Arduino IDE, (but not the new version 2.x). IT helps to know that the module is basically the WisDuo Module RAK3172-SiP/RAK3172LP-SiP which in the board manager you need to choose the RAK3272-SiP board
. Connecting to the sticker looks do-able.
Helium connectivity with TrackPac.io is hard to test until my QR code arrives. Please like if you are also interested in the RAK2270 Sticker Tracker
1 Like
Hi Jeremy,
I am also giving answers in github and linkedin in all your questions 
As I said, the accelerometer is based on RAK1904 so our software guide for that module can be a starting point.
For machine learning, you can probably have a look on Edge Impulse but that is something you have to host somewhere (not edge deployment). Edge deployment is only possible on RAK4630, RAK11200 and RAK11300 modules.
For LoRaWAN, you can consider the modular example of @beegee 's RUI3 Sensor Node examples. This is complete package and uses cayenneLPP based decoder.
You can also have a look on our WisBlock Kit 3 guide since RAK1904 is used there. You can ignore the GPS part. It uses a direct decoder with some bitwise manipulation to fit data on group of byte-size data. This is lean and efficient but not as portable as CayenneLPP based decoders.
1 Like
Can you expand on this @carlrowan I am part of the Edgeimpulse.com expert network, as long as the accelerometer works with the Arduino IDE we should be able to get a small ML model working on the device.
I see the RAK documentation Getting Started with WisBlock and Edge Impulse | RAKwireless Documentation Center
The RAK2270 Sticker Tracker is STM32 based I think it should be able to run an ML model. Most of my work has been with the Arduino PortentaH7 which is also STM32 based.
Hi Jeremy,
Tiny ML models can run on STM32 chips like what is inside RAK2270. However, this requires energy (will drain the thin battery very fast). Just like I said on my message, what we can do is maybe create models on the data generated by the RAK Sticker in longer time scale and perform the ML algorithms externally. If there is enough pattern on it, surely we can run a prediction model or other ML algorithms.
I did something like this before. A Raspberry Pi gets the LoRaWAN payload from the network server then performs the prediction algorithm. I used this forecasting algorithm - Time Series Forecasting with Supervised Machine Learning | by Unai López Ansoleaga | Towards Data Science
This ensures that the LoRaWAN device powered by battery won’t need to consume energy on crunching numbers and arrays 
1 Like
I was thinking of taking the temperature every 10 mins for an entire day and then doing some machine learning or just data monitoring to determine how stable the temperature was during the day and then sending the data once a day over LoRaWan.
Once I can program the accelerometer something similar would also be useful.