Low-Power, Low-Latency On-Device AI Vision Integrated Solution
In edge‑AI domains such as industrial vision, autonomous robotics, intelligent security, and embedded quality inspection, the industry has long grappled with three major pain points: excessive power consumption in high‑performance computing hardware, latency and stuttering in multi‑channel video inference, and insufficient native computational capacity on embedded motherboards to handle complex, concurrent multi‑model workloads. And… XC3588E Industrial Control Motherboard Pairing DEEPX DX-M1 Acceleration Card a combined solution, implemented through a hierarchical computing architecture Ultra-low overall power consumption, millisecond-level inference latency, and multi-channel concurrent vision processing. Three breakthroughs deliver a one-stop, high-performance visual computing foundation for lightweight edge‑side intelligent devices.
XC3588E Motherboard: An Embedded Computing Platform Custom‑Designed for DEEPX
XC3588E It features a Rockchip RK3588 octa-core processor built on an 8nm process (4×A76 high-performance cores + 4×A55 low-power efficiency cores), with a native 6 TOPS NPU that handles core system scheduling, 8K video encoding and decoding, peripheral interfacing, multi-sensor data fusion, and general-purpose computational tasks. Lightweight AI inference can run locally on its own, while standby power consumption remains well‑controlled.
XC3588E features deep-level adaptation across hardware interfaces, power supply, and the system layer. DEEPX DX-M1 accelerator card: no riser card required, no additional modifications—ready to deploy right out of the box.
- Native support for the standard M.2 M-KEY slot : Dedicated onboard PCIe slot, optimized power delivery circuitry, stable power supply DEEPX Low-power operation;
- Comprehensive coverage of industrial peripheral interfaces : 9 USB ports, 6 serial ports, dual Gigabit Ethernet ports, dual CAN buses, multiple GPIOs, and EDP/MIPI/HDMI/LVDS display interfaces; it can simultaneously connect industrial cameras, LiDAR sensors, servo motors, audio‑visual alarms, and local displays, offering a one‑stop integration of vision and motion control.
- Multi-system compatibility : Compatible Android, Ubuntu, and the domestically developed Kylin operating system , Zero‑barrier development enables rapid deployment of diverse edge‑side AI vision projects.

Soft-Hard Hierarchical Computing Architecture: Each component performs its designated role, achieving extreme power efficiency.
DEEPX DX-M1 Acceleration Card By leveraging the motherboard’s standard M.2 M-KEY PCIe 3.0 ×4 high-speed interface for expansion, it is optimized to handle high‑load, multi‑threaded concurrent deep learning inference, thereby forming… CPU + onboard NPU + discrete visual acceleration card Tier-3 computing power division:
- RK3588 main controller : Integrates the entire system, image acquisition, device interoperation, and local data storage;
- RK3588 features a built-in 6 TOPS NPU. : Lightweight face detection and single-channel simple object recognition;
- DEEPX Independent Acceleration Unit : Supports high‑load vision models such as YOLOv8/v9, semantic segmentation, pose estimation, 3D point clouds, and multi‑camera synchronized detection.
DEEPX DX-M1 The core advantage lies in 25 TOPS INT8 peak computing performance, with the entire system’s accelerated power consumption ranging from only 2.69 W to 5 W. The single card comes with 4 GB of dedicated LPDDR5 cache, ensuring that inference memory is entirely independent and does not consume motherboard memory resources. With no redundant computing power stacking, its big‑core and small‑core dynamic frequency scaling, combined with the accelerator card’s intelligent start‑stop functionality, enables precise three‑stage power management—standby, idle, and full load—significantly extending battery life for mobile robots and power‑supply inspection devices.
Multi-industry implementation scenarios
1. Autonomous Mobile Robots (Warehouse AGVs, Inspection Robots, Collaborative Robotic Arms)
DEEPX Millisecond‑level visual reasoning enables real‑time obstacle detection, storage‑location verification, and human‑obstacle avoidance. The RK3588 simultaneously handles radar‑based SLAM mapping and motion control, supporting stable 24/7 offline operation for outdoor power‑line inspection and unmanned forklifts in manufacturing workshops.

2. Industrial Vision Online Quality Inspection
Synchronous access to multiple industrial cameras, DEEPX It simultaneously detects solder joint defects, surface imperfections, and dimensional discrepancies, delivering millisecond‑level inspection results that trigger an integrated rejection mechanism. The system’s low‑power design is compatible with enclosed production line control cabinets, eliminating the need for high‑capacity cooling equipment and reducing workshop operation and maintenance costs.

3. Smart Security and Smart Construction Sites
Multi-camera, full‑area real‑time analytics with local, real‑time alerts—no cloud forwarding required. Its low‑power design is optimized for PV‑powered outdoor monitoring devices; even when disconnected from the network, it continues AI‑driven local inference, while locally stored data ensures privacy and security.

4. Embedded Intelligent Terminals (Self-Service Devices, In-Vehicle Edge Boxes, Medical Vision Terminals)
For compact, power-constrained edge devices, this small‑form‑factor, low‑power solution enables on‑device facial verification, product recognition, and preliminary screening of medical images—delivering smooth, lag‑free performance while mitigating cloud‑based network latency and data‑exposure risks.

The XC3588E industrial‑grade motherboard paired with the DEEPX DX‑M1 accelerator card leverages a hierarchical, collaborative computing architecture. Built on domestically developed AI‑enabled hardware and enhanced by specialized visual‑processing acceleration, this solution delivers a cost‑effective, production‑ready edge‑side AI integration platform for industrial automation, intelligent robotics, edge‑based visual monitoring, and other applications—enabling lightweight, on‑premises AI upgrades across diverse industries.
Next page
Recommended
From data center infrastructure monitoring to intelligent manufacturing, from energy power plants to AI‑driven security, in the era of domestically developed edge computing, Xiangcheng Electronics’ industrial control host series provides a stable and reliable hardware foundation, helping industries across the board achieve digital and intelligent transformation.
The XC3576H industrial control motherboard has now been fully adapted to Ubuntu 26.04 LTS, offering developers a high-performance, long‑support‑cycle, and feature‑rich ARM‑based open‑source solution for industrial applications.
The IPCA_3588E industrial control host features a self-developed XC3588E industrial motherboard and is built with exclusively new, original‑brand chips, delivering an integrated edge computing terminal that combines domestically produced chips with a domestic operating system.

