TI-RTOS-MCU: Machine Learning on CC32XX

Pablo Munoz

Prodigy 180 points

Part Number: TI-RTOS-MCU

Is there any tutorial/guide/information on how to deploy a ML model in a CC3220 ?

the process goes like this:

CC3220 receives 10 readings from different type of sensors
the readings are passed to the ML inference function
blink a led depending on the value returned from ML inference function

I'm using TI RTOS.

Any libraries I can use?

I've seen Tensorflow Lite, but I'm not sure it can be used from TI RTOS. Also it is built on C++

Any guidance will be very much appreciated

over 3 years ago

0 Pablo Munoz over 3 years ago

Prodigy 180 points

I found this TI Session [1] where they do exactly what I'm looking for. And I would like to get more information on:

How to build, deploy and use the C++ library on a launchpad with TI RTOS
Would it work on a CC3220 with an M4 Cortex as mentioned in the video? Or I absolutely need an M4F to take advantage of the FPU?
Is TI going to produce any solution around this approach? thinking about hundreds or thousands of sensors this process needs to be automated.

[1] Enabling low-power industrial wireless sensors with deep learning capabilities for audio, sound and voice, Nov 2020. https://training.ti.com/enabling-low-power-industrial-wireless-sensors-deep-learning-capabilities-audio-sound-and-voice

thanks

0 Jesu over 3 years ago in reply to Pablo Munoz

TI__Mastermind 22935 points

Hi Pablo,

I'm not familiar with any solution like this available for the CC32XX. I do know we support C++ on the CC32XX though. Check this link. Mathematically speaking, I don't see why you cannot implement a neural network on a CC32XX but I would imagine it would take a lot of CPU cycles without the FPU like on the M4F.

Jesu

Arm-based microcontrollers

Arm-based microcontrollers forum

TI-RTOS-MCU: Machine Learning on CC32XX