High-Performance AI Server for Autonomous Sorting Line
12× NVIDIA Jetson Orin AGX Cluster for Real-Time Warehouse Automation
We developed an industrial AI server that controls a sorting line and a fleet of autonomous robots in real time. The system integrates 12 NVIDIA Jetson Orin AGX modules in a cluster, redundant power, and a controller — enabling high-speed local video processing, navigation, and sorting without cloud dependency.
Project Summary
Autonomous sorting line control
12× NVIDIA Jetson Orin AGX compute modules in a cluster
Low-latency vision + robotics coordination
Redundant power subsystem (24/7 uptime)
Full-cycle R&D
In-house laboratory
100% Job Success
(Upwork & Linkedin)
(Upwork & Linkedin)
Project Overview
Client: Multi-story Automated sorting facility
Objective: Build a real-time compute cluster for controlling robotics and sorting line operations
Platforms: NVIDIA Jetson Orin AGX / Xavier AGX
Timeline: From architecture to prototype and pilot production
Our contribution: System architecture, PCB design, embedded software, Linux BSP, drivers
Objective: Build a real-time compute cluster for controlling robotics and sorting line operations
Platforms: NVIDIA Jetson Orin AGX / Xavier AGX
Timeline: From architecture to prototype and pilot production
Our contribution: System architecture, PCB design, embedded software, Linux BSP, drivers
The client was building a fully autonomous sorting line where dozens of mobile robots and conveyor modules must operate in perfect synchronization. This required a local high-performance computing center with low latency, industrial reliability, and full autonomy from cloud services.
We developed a custom scalable AI server based on 12 NVIDIA Jetson Orin AGX modules, combining them into a unified industrial compute cluster.
We developed a custom scalable AI server based on 12 NVIDIA Jetson Orin AGX modules, combining them into a unified industrial compute cluster.
Engineering Challenges
High-performance scalable cluster of 12 Jetson Orin AGX modules
Stable execution of parallel workloads: video pipelines, navigation, sorting logic.
Ultra-low latency for robot control
All decisions and commands must be issued instantly — no cloud, no network delays.
Redundant power subsystem
Reliable 24/7 operation under heavy load and fluctuating line throughput.
Industrial-grade cooling
Thermal stability during continuous GPU/CPU peak loads.
Integration with industrial line controller
Syncing robots, conveyors, scanners, triggers, and sensors.
Dense high-complexity PCB
1200+ components, 3800+ vias, PCIe Gen4, USB 3.1, M.2, multi-rail power.
How the Server Powers the Sorting Line
The server acts as the central brain of the entire system — coordinating robots, conveyors, scanners, and tracking all product flows.
It performs:
▪️ real-time video processing,
▪️ object and barcode recognition,
▪️ sorting and routing decisions,
▪️ robot path planning and optimization,
▪️ synchronization of all moving components,
▪️ fully local work with zero cloud dependency.
This transforms the sorting floor into a fully autonomous, high-throughput system.
▪️ real-time video processing,
▪️ object and barcode recognition,
▪️ sorting and routing decisions,
▪️ robot path planning and optimization,
▪️ synchronization of all moving components,
▪️ fully local work with zero cloud dependency.
This transforms the sorting floor into a fully autonomous, high-throughput system.
Results
▪️ Sorting throughput increased
▪️ Decision-making latency significantly reduced
▪️ Stable industrial 24/7 operation under high load
▪️ System easily scales by adding more Jetson modules
▪️ Full independence from external servers and internet
▪️ Decision-making latency significantly reduced
▪️ Stable industrial 24/7 operation under high load
▪️ System easily scales by adding more Jetson modules
▪️ Full independence from external servers and internet
Key Parameters
▪️ SoM: NVIDIA Jetson Orin AGX / Xavier AGX
▪️ Interfaces: USB 3.1, 1G Ethernet
▪️ Expansion: M.2 SATA / PCIe Gen4 x4
▪️ PCB: 8 layers, 1200+ components
▪️ Complexity: 4300+ pads, 3800+ vias
▪️ Interfaces: USB 3.1, 1G Ethernet
▪️ Expansion: M.2 SATA / PCIe Gen4 x4
▪️ PCB: 8 layers, 1200+ components
▪️ Complexity: 4300+ pads, 3800+ vias