Elementor #4036 – Grovety

We implement ML on microcontrollers

Making trained neural networks fit for inference
on microcontrollers

That is limited by:

At the same time:

What we do:

Features:

Solving tasks of neural network learning and fine-tuning, transfer learning on microcontrollers
(On Device Learning, ODL)

GRC AI module

A self-learning engine doing Anomaly Detection and n-Class Classification in the time series data flow for providing a full turnkey AI functionality for integration in products of various complexity. AI Module is based on the ODL (On-Device Learning) computing approach for providing and enhancing AI scenarios.

GRC AI Module is an innovative and user-friendly solution for introducing AI functionality to your products in effortless and time-efficient manner. AI-based data processing offers a more flexible and powerful approach to analyzing and making decisions from large data collection. Machine learning algorithms autonomously learn patterns and relationships from the time series data flow.

The GRC AI Module is

A pre-installed firmware with neural network based on the next-generation ODL Computing approach. It provides anomaly detection and classification of input signals in accordance with the pre-trained data.
Intended for the use on ESP32 C3 FN4/H4 MCU, an optimal choice for AI-related scenarios and Artificial Intelligence of Things (AIoT) due to the acceleration for neural network computing and signal processing workloads provided by the SoC.

GRC AI-module

Development board

Training and implementation of neural network is performed on ESP32-C3

The performance of an on-board neural network is done on the development board microcontroller by processing input data gathered from either connected or embedded sensors and learning on-the-fly. You do not have to be an AI expert, prepare a dataset, or train a model. AI functionality is available «out of the box».

Dev Board User Guide

Dev Board Developer’s Guide (including code examples)

GRC Developer's Guide (including recommendations on embedding AI Module into other devices);

Comprehensive consultation in Edge AI questions:

Technologies

ML compilers:

TFL+TFLµ; TVM+µTVM

SDK for inference:

All available tools that simplify and speed up the design of final solutions

Neural Network Optimization

For trained NNs

Compiler and hardware level

Runtime

Heterogeneous microcontroller architectures with NNA

On Device Learning (ODL)

ML compilers:

The goal is to create more efficient smarter devices by figuring out changes in the current dataflow, so devices would autonomously adjust or reconfigure their operating model. If necessary, the “knowledge” gained by one device is shared with other connected devices.