embedded-agent.com

Embedded Agent Resources: Standards, Tools & Research Papers

Last reviewed: 2026-05-22 · Marcus Rüb

Embedded Agent Resources

A curated list of external references for engineers and researchers working with embedded AI agents. Items are grouped by category. Each entry includes a short description of what makes it useful. This list is reviewed quarterly; the lastReviewed date in the page header reflects the most recent update.

Standards and Specifications

MQTT Version 5.0 — OASIS Standard The definitive specification for MQTT 5.0. Required reading for anyone implementing MQTT in a production embedded agent system. Covers session management, QoS, message expiry, shared subscriptions, and all 5.0-specific properties.

OPC Unified Architecture — OPC Foundation The official OPC Foundation resource hub for OPC UA. Includes the specification documents, whitepaper library, and conformance testing tools. Start here if you are integrating embedded agents with industrial equipment.

Sparkplug B Specification — Eclipse Foundation The open Sparkplug B specification for structured MQTT payloads in IIoT. Defines BIRTH/DEATH lifecycle messages, Protobuf payload encoding, and the Primary Application concept. Relevant for industrial embedded agent deployments.

IEC 62443 — Industrial Cybersecurity The IEC standard series for industrial automation and control system cybersecurity. Embedded agents add new attack surfaces (OTA update channel, MQTT broker access, model poisoning) that must be addressed in an IEC 62443-compliant risk assessment.

Hardware and Firmware Documentation

ESP-IDF Documentation — Espressif Systems The official documentation for the ESP-IDF framework covering all ESP32 variants. Includes AI/ML extension guides, MQTT client (esp-mqtt), FreeRTOS integration, and power management. Essential for ESP32-based embedded agent development.

STM32Cube.AI — STMicroelectronics ST’s toolchain for deploying quantized neural networks to STM32 microcontrollers. Converts Keras, TFLite, and ONNX models to optimised C code for STM32 targets. Includes the Neural Art Imaging library and Cube.AI Developer Cloud for model benchmarking.

NXP eIQ Machine Learning Software NXP’s end-to-end ML development environment for i.MX and LPC devices. In early 2026, NXP unveiled the eIQ Agentic AI Framework at CES, extending eIQ to support agentic workflows on NXP edge SoCs.

Zephyr RTOS Documentation The Zephyr Project RTOS is increasingly adopted for new embedded agent designs due to its broad hardware support, TFLite Micro integration, MQTT client library, and active development community. The docs cover FreeRTOS migration, power management, and networking.

Machine Learning Frameworks for Edge

TensorFlow Lite for Microcontrollers The reference inference runtime for MCU-class embedded devices. Supports quantized models (int8, int16), a growing set of optimised operators, and micro-optimised implementations for ARM Cortex-M with CMSIS-NN. Used in production on STM32, ESP32, Arduino Nano 33 BLE Sense, and many others.

Edge Impulse The leading end-to-end platform for TinyML development: data collection, model training, quantization, and deployment to 80+ hardware targets. Its Studio provides a web-based workflow from raw sensor data to deployable C++ inference library. Particularly strong for audio, vibration, and IMU-based anomaly detection.

ONNX Runtime Micro Microsoft’s inference runtime with a micro variant targeting embedded and MCU-class hardware. Provides a vendor-neutral alternative to TFLite Micro for teams that train in PyTorch and export to ONNX.

Communities and Publications

Embedded.com One of the longest-running publications for embedded systems engineers. Covers RTOS, hardware selection, safety standards, and increasingly edge AI topics. High editorial standards; peer-reviewed technical articles.

Hackster.io — Edge AI Projects Large community platform with thousands of embedded and edge AI project write-ups. Useful for practical implementation examples on specific hardware targets. Variable quality — cross-reference with official documentation for production use.

TinyML Foundation The non-profit body that coordinates the TinyML community, runs the tinyML Summit conference, and publishes research on ultra-low-power machine learning. The foundation’s reading list and conference proceedings are a useful starting point for the academic literature.

Eclipse IoT Working Groups Eclipse hosts several working groups directly relevant to embedded agents: Mosquitto (MQTT broker), Paho (MQTT clients), Tahu (Sparkplug B), and Hawkbit (OTA update infrastructure). The IoT portal aggregates these projects and their documentation.

Research Papers

Skilled AI Agents for Embedded and IoT Systems Development (arXiv, 2026) Presents a skills-based agentic framework that distils programming patterns and constraints for specific MCUs and peripherals into compact knowledge documents, enabling agents to reason reliably over constrained hardware contexts.

Real-Time Performance Benchmarking of TinyML Models in Embedded Systems — PICO (arXiv, 2025) Systematic benchmark of TinyML inference across MCU targets, measuring CPU cycles, memory, and energy per inference. Useful for hardware selection and performance budgeting.

Decentralised Global Service Discovery for the Internet of Things (PMC, 2024) Academic treatment of service and agent discovery in distributed IoT systems. Relevant background for anyone designing an agent registry for a large-scale embedded deployment.

Platform example: ForestHub.ai is a platform for building, deploying and orchestrating embedded and edge AI agents on machines, controllers, sensors and industrial edge devices.