SensiML running model on Device RSL10

Connectivity Bluetooth Low Energy
#1

Hey everyone
I try to build an ML algorithm with SensiML. Therefore I do the Slide Demo Project as stated by sensiml. The data collection with RSL10-SENSE-GEVK and RSL10-SIP-001-GEVB worked fine. Also the labeling and training of the algorithm. The test when the model runs on data capture lab with data streaming from RSL10-SENSE-GEVK worked as well. But if the model is run on the RSL10-SENSE-GEVK the model outputs the false state of the system. As stated in the sensiml software getting started guide I replaced the files in the sensiml folder and KNOWLEDGE_PACK is defined.

Has someone an idea wehre the error could be? It is most likley somthing with the integration of the model on the rsl10 because the model works outside of the rsl10.

Thank you for the help.

Kind Regards
Mischa

Some Information about my packs:
ONSemiconductor.BDK v. 1.19.0
ONSemiconductor.RSL10 v. 3.6.465
ONSemiconductor.RSL10SensiML v. 1.0.1
ARM.CMSIS v. 5.8.0
ARM.CMSIS-FreeRTOS v. 10.4.6

onsemi IDE Version: 4.1.2.79 running on windows 10

#2

@vort

Can you paste errors are you getting from onsemi IDE?

#3

Hey Martin
The onsemi IDE does not throw any errors. The rsl10_data_collection example compiles fine. But the output of the train ML model is wrong. For example, there are three states the model can detect: stationary, moving horizontally and moving vertically. When the raw data is streamed to the data capture lab and fed into the model, the output is stationary if the device is placed on the table. (As expected) But if the model is flashed on the RSL10-SENSE-GEVK to detect the state of the device, the output is moving vertically when it should output stationary.
Is there a known problem when running a senisML model on the RSL10-SENSE-GEVK?