ROKOMARI IT
Edge Intelligence & Automation Modules

EdgeBrain AI Module

On-board ML inference — no cloud, no latency, no limits.

Device Status

Operational Features

  • ATEX intrinsically safe certifiable design
  • DIN-rail IP66 or IP67 enclosure grade
  • RS485 Modbus RTU telemetry output built-in
  • mTLS secure transport authentication ready
Introduction

Product Overview

Compact, high-efficiency compute expansion board designed for rapid custom board integration to execute on-site ML modeling.

Operational Scope & Logic

Rockchip RK3566 quad-core Cortex-A55 with integrated 1 TOPS NPU. Designed as an M.2/board-to-board expansion for edge MCU systems. Runs TFLite, ONNX, and YOLO models natively.

Profile Specifications

CPU Quad-core ARM Cortex-A55 @ 1.8 GHz
NPU 1 TOPS (INT8)
RAM 4 GB LPDDR4
Storage 32 GB eMMC
OS Ubuntu 22.04 / Buildroot Linux
Key Deliverables

Core Capabilities

Rockchip RK3566 quad-core A55 @ 1.8 GHz
Integrated 1 TOPS Neural Processing Unit
Runs TFLite / ONNX / YOLOv8 natively
PCIe + USB 3.0 + MIPI-CSI expansion interfaces
Fanless passive-cooled design
Field Integrations

Applied Use Cases

  • Plug-in vision AI for factory conveyor camera systems
  • On-board predictive maintenance inference for motors
  • NLP/LLM inference for LingoControl LLM Middleware
Interactive Sandbox

Hardware & API Simulator

Test telemetry registers, compilation sweeps, or load parameters live using our built-in R&D diagnostic simulator below.

R&D Sandbox Tool

Post-Training Quantization Compiler

Compile high-dimensional neural network weights (FP32) into compact micro-instructions (INT8).

Original Model (FP32) 96.4 MB
Quantized Target (INT8) -- MB
Ready to initialize compiler loop...
Operational Integrity

Ready for Immediate Factory Deployment & Custom Integration

Our design processes follow strict quality management pipelines, ensuring physical hardware modules fit seamlessly inside dynamic industrial cabinets.