Hey everyone,
I’ve been experimenting with the Seeed Studio reComputer Mini J3011 for an edge AI computer vision project and wanted to share some notes that might be useful for others considering this platform.
What it is (briefly)
The Mini J3011 is a tiny form factor edge AI box measuring just 63mm × 95mm × 42mm — built around the NVIDIA Jetson Orin Nano 8GB module, delivering up to 40 TOPS of AI performance via the Ampere GPU architecture and a 6-core ARM CPU. The “with extension” variant adds a bottom PCIe board that unlocks up to 8 USB ports and dual CAN bus, which is handy for robotics use cases.
Getting started
Good news: it ships with JetPack 6.0 pre-installed on a 128GB NVMe SSD, so you’re not flashing from scratch. Out of the box you get CUDA, cuDNN, and TensorRT ready to go.
For computer vision inference I tested YOLOv8 via Ultralytics, which works well. The Seeed wiki has a Getting Started with reComputer page that covers JetPack flashing if you ever need to recover or upgrade.
Power + deployment note
One thing that stood out: it supports direct 54V DC input, making it well suited for battery-powered systems like drones or patrol robots. If you’re embedding this into a rover or AMR, that’s a real advantage over units that need a 12V regulated rail.
Compared to a standard Jetson Orin Nano Dev Kit
The dev kit is easier to prototype on (more exposed I/O). The Mini J3011 is more suited to deployment — compact, enclosed, and supports OTA updates and remote fleet management via Allxon and Balena, which is useful once you have more than a couple of units in the field.
Questions for the community
- Has anyone used the dual CAN interface on the extension board with ROS2? Curious about latency for motor control loops.
- Any experience running Vision-Language Models (e.g. LLaVA) on the 8GB variant? I’m right at the edge of VRAM with some quantised models.
Happy to answer questions from anyone just getting started with Jetson Orin Nano. Still learning myself but glad to compare notes.
