LiteRT para microcontroladores se diseñó para ejecutar modelos de aprendizaje automático
en microcontroladores y otros dispositivos
con solo unos pocos kilobytes de memoria. El
del entorno de ejecución principal cabe en 16 KB en un ARM Cortex M3 y puede ejecutar muchas
e implementar modelos automáticamente. No requiere compatibilidad con el sistema operativo, tipos C o C++ estándar
bibliotecas o asignación de memoria dinámica.
Por qué son importantes los microcontroladores
Por lo general, los microcontroladores son dispositivos
de procesamiento pequeños y de baja potencia
incorporadas en hardware que requieren procesamiento básico. Si llevamos el aprendizaje
a microcontroladores muy pequeños, podemos potenciar la inteligencia de miles de millones
de dispositivos que usamos en la vida, como los electrodomésticos y la Internet de
Usa diferentes dispositivos sin depender de hardware costoso ni de una conexión a Internet confiable
de red, que a menudo está sujeto a restricciones de ancho de banda y energía, y
genera una latencia alta. Esto también puede ayudar a preservar la privacidad, ya que no hay datos
sale del dispositivo. Imagina electrodomésticos inteligentes que puedan adaptarse a tu día a día
con sensores industriales inteligentes
y rutinarios que comprenden la diferencia
problemas y el funcionamiento normal, y juguetes mágicos que pueden ayudar a los niños a aprender de manera divertida.
de formas encantadoras.
Plataformas compatibles
LiteRT para microcontroladores está escrito en C++ 17 y requiere una arquitectura
plataforma. Se probó ampliamente con muchos procesadores basados en
Arm Cortex-M Series
y se trasladó a otras arquitecturas, como
ESP32 El
está disponible como una biblioteca de Arduino. También puede generar proyectos para
de desarrollo de software, como Mbed. Es de código abierto y se puede incluir en
cualquier proyecto de C++ 17.
Cada aplicación de ejemplo está en
GitHub
y tiene un archivo README.md que explica cómo se puede implementar en sus
y plataformas de Google Cloud. Algunos ejemplos también tienen tutoriales de extremo a extremo con un
plataforma como se indica a continuación:
Hello World:
Demuestra los conceptos básicos del uso de LiteRT para
Microcontroladores
Convierte a un array de bytes en C con
herramientas estándar
memoria de programa de solo lectura en el dispositivo.
Ejecuta la inferencia en el dispositivo con el proceso y la biblioteca de C++.
los resultados.
Limitaciones
LiteRT para microcontroladores se diseñó para cumplir con las restricciones específicas de
y el desarrollo de microcontroladores. Si trabajas en dispositivos más potentes (por ejemplo,
ejemplo, un dispositivo Linux incorporado como Raspberry Pi), el estándar
Es posible que el framework LiteRT sea más fácil de integrar.
Debes tener en cuenta las siguientes limitaciones:
[[["Fácil de comprender","easyToUnderstand","thumb-up"],["Resolvió mi problema","solvedMyProblem","thumb-up"],["Otro","otherUp","thumb-up"]],[["Falta la información que necesito","missingTheInformationINeed","thumb-down"],["Muy complicado o demasiados pasos","tooComplicatedTooManySteps","thumb-down"],["Desactualizado","outOfDate","thumb-down"],["Problema de traducción","translationIssue","thumb-down"],["Problema con las muestras o los códigos","samplesCodeIssue","thumb-down"],["Otro","otherDown","thumb-down"]],["Última actualización: 2025-07-24 (UTC)"],[],[],null,["# LiteRT for Microcontrollers is designed to run machine learning models\non microcontrollers and other devices with only a few kilobytes of memory. The\ncore runtime just fits in 16 KB on an Arm Cortex M3 and can run many basic\nmodels. It doesn't require operating system support, any standard C or C++\nlibraries, or dynamic memory allocation.\n| **Note:** The [LiteRT for Microcontrollers Experiments](https://experiments.withgoogle.com/collection/tfliteformicrocontrollers) features work by developers combining Arduino and TensorFlow to create awesome experiences and tools. Check out the site for inspiration to create your own TinyML projects.\n\nWhy microcontrollers are important\n----------------------------------\n\nMicrocontrollers are typically small, low-powered computing devices that are\nembedded within hardware that requires basic computation. By bringing machine\nlearning to tiny microcontrollers, we can boost the intelligence of billions of\ndevices that we use in our lives, including household appliances and Internet of\nThings devices, without relying on expensive hardware or reliable internet\nconnections, which is often subject to bandwidth and power constraints and\nresults in high latency. This can also help preserve privacy, since no data\nleaves the device. Imagine smart appliances that can adapt to your daily\nroutine, intelligent industrial sensors that understand the difference between\nproblems and normal operation, and magical toys that can help kids learn in fun\nand delightful ways.\n\nSupported platforms\n-------------------\n\nLiteRT for Microcontrollers is written in C++ 17 and requires a 32-bit\nplatform. It has been tested extensively with many processors based on the\n[Arm Cortex-M Series](https://developer.arm.com/ip-products/processors/cortex-m)\narchitecture, and has been ported to other architectures including\n[ESP32](https://www.espressif.com/en/products/hardware/esp32/overview). The\nframework is available as an Arduino library. It can also generate projects for\ndevelopment environments such as Mbed. It is open source and can be included in\nany C++ 17 project.\n\nThe following development boards are supported:\n\n- [Arduino Nano 33 BLE Sense](https://store-usa.arduino.cc/products/arduino-nano-33-ble-sense-with-headers)\n- [SparkFun Edge](https://www.sparkfun.com/products/15170)\n- [STM32F746 Discovery kit](https://www.st.com/en/evaluation-tools/32f746gdiscovery.html)\n- [Adafruit EdgeBadge](https://www.adafruit.com/product/4400)\n- [Adafruit LiteRT for Microcontrollers Kit](https://www.adafruit.com/product/4317)\n- [Adafruit Circuit Playground Bluefruit](https://learn.adafruit.com/tensorflow-lite-for-circuit-playground-bluefruit-quickstart?view=all)\n- [Espressif ESP32-DevKitC](https://www.espressif.com/en/products/hardware/esp32-devkitc/overview)\n- [Espressif ESP-EYE](https://www.espressif.com/en/products/hardware/esp-eye/overview)\n- [Wio Terminal: ATSAMD51](https://www.seeedstudio.com/Wio-Terminal-p-4509.html)\n- [Himax WE-I Plus EVB Endpoint AI Development Board](https://www.sparkfun.com/products/17256)\n- [Synopsys DesignWare ARC EM Software Development Platform](https://www.synopsys.com/dw/ipdir.php?ds=arc-em-software-development-platform)\n- [Sony Spresense](https://developer.sony.com/develop/spresense/)\n\nExplore the examples\n--------------------\n\nEach example application is on\n[GitHub](https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/examples)\nand has a `README.md` file that explains how it can be deployed to its supported\nplatforms. Some examples also have end-to-end tutorials using a specific\nplatform, as given below:\n\n- [Hello World](https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/examples/hello_world) - Demonstrates the absolute basics of using LiteRT for Microcontrollers\n - [Tutorial using any supported device](./get_started)\n- [Micro speech](https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/examples/micro_speech) - Captures audio with a microphone to detect the words \"yes\" and \"no\"\n - [Tutorial using SparkFun Edge](https://codelabs.developers.google.com/codelabs/sparkfun-tensorflow/#0)\n- [Person detection](https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/examples/person_detection) - Captures camera data with an image sensor to detect the presence or absence of a person\n\nWorkflow\n--------\n\nThe following steps are required to deploy and run a TensorFlow model on a\nmicrocontroller:\n\n1. **Train a model** :\n - *Generate a small TensorFlow model* that can fit your target device and contains [supported operations](./build_convert#operation_support).\n - *Convert to a LiteRT model* using the [LiteRT converter](./build_convert#model_conversion).\n - *Convert to a C byte array* using [standard tools](./build_convert#convert_to_a_c_array) to store it in a read-only program memory on device.\n2. **Run inference** on device using the [C++ library](./library) and process the results.\n\nLimitations\n-----------\n\nLiteRT for Microcontrollers is designed for the specific constraints of\nmicrocontroller development. If you are working on more powerful devices (for\nexample, an embedded Linux device like the Raspberry Pi), the standard\nLiteRT framework might be easier to integrate.\n\nThe following limitations should be considered:\n\n- Support for a [limited subset](./build_convert#operation_support) of TensorFlow operations\n- Support for a limited set of devices\n- Low-level C++ API requiring manual memory management\n- On device training is not supported\n\nNext steps\n----------\n\n- [Get started with microcontrollers](./get_started) to try the example application and learn how to use the API.\n- [Understand the C++ library](./library) to learn how to use the library in your own project.\n- [Build and convert models](./build_convert) to learn more about training and converting models for deployment on microcontrollers."]]
