Brains, eyes and hands: building AI robots and automation, from DIY to production
Robots and automated machines used to need a big budget and a specialist team. That has changed. Affordable AI computers, depth cameras and open-source robot arms now let a student, a startup or a factory team build something that sees, decides and acts, and then grow it into a real deployment.
Every one of these systems has the same three parts:
| Part | What it does | What we stock |
|---|---|---|
| Brain | Runs the AI that understands what the machine sees and decides what to do | NVIDIA Jetson developer kits, modules and ready-built edge computers |
| Eyes | Gives the machine sight and a sense of depth | Stereolabs ZED stereo and single cameras |
| Hands | Acts in the physical world | Seeed reBot robot arms |
Pick one of each, and you have a working system.
The brain: NVIDIA Jetson
Jetson is NVIDIA's family of small, power-efficient AI computers built to sit inside machines rather than in a data centre. They all run the same NVIDIA software, so what you build on a developer kit carries over to production hardware.
- Jetson Orin Nano Super Developer Kit: the affordable place to start learning and prototyping.
- Jetson AGX Orin 64GB Developer Kit: for heavier vision work and several cameras at once.
- Jetson AGX Thor Developer Kit: built on NVIDIA Blackwell for advanced robots, humanoids and physical AI.
The eyes: Stereolabs ZED cameras
A normal camera sees a flat picture. A stereo camera works like a pair of eyes: it measures how far away things are, so a machine can avoid obstacles, pick up objects and map a room.
- USB cameras (ZED Mini, ZED 2i) plug straight into a laptop or developer kit, so they're the easy way to start.
- GMSL2 cameras (ZED X, ZED X Mini) are rugged and sealed against dust and water, and they support longer cables and several cameras in sync. They're built for robots and machines that work outdoors or on a factory floor.
- Single cameras (ZED X One range) cover surround vision, or high-detail inspection where you need resolution rather than depth.
The hands: Seeed reBot arms
The reBot arms are fully open source: hardware, software, design files and parts list. They work with ROS, NVIDIA Isaac Sim and LeRobot, the tools most robot learning projects use.
- reBot Arm B601-RS is assembled and ready to use, with a 2.5kg rated payload and ±0.1mm repeatability.
- reBot Arm B601-DM is the assembled 24V version.
- B601-RS unassembled kit is for teams who want to build and understand the hardware themselves. It suits universities and STEM programs well.
- Star Arm 102 leader arm lets you move the B601-DM by hand. The robot copies you, and you can record those movements to teach an AI model the task. This is how many robot learning projects collect their training data.
Three ways to build
1. DIY and learning
For: students, makers, educators, and engineers exploring robot learning.
A starter setup: a Jetson Orin Nano Super Developer Kit, a ZED Mini or ZED 2i, and a reBot arm with the Star Arm leader. Start in simulation with Isaac Sim, then move to the real arm. Record demonstrations with the leader arm and train the robot to repeat a task such as picking and placing objects.
2. Automation and inspection
For: manufacturers, warehouses, agriculture and logistics teams that want machines to spot defects, count items, read labels or guide equipment.
A typical setup: a ready-built edge computer such as a Seeed reComputer or an AVerMedia box PC, with rugged ZED X cameras. For several cameras on one system, the AVerMedia GMSL module connects up to eight. The AI runs on site, so footage stays on your premises and decisions happen in real time without depending on an internet connection.
3. Advanced robotics and physical AI
For: robotics startups and research labs building mobile robots, humanoids and autonomous machines.
A typical setup: the Jetson AGX Thor Developer Kit for development, moving to a Jetson Thor module in your own design, or a deployment-ready system such as the Connect Tech Anvil-T5 or AVerMedia D331ATB. Add Seeed reComputer Robotics units for robots that need many camera inputs and industrial connections.
Where it's used in the real world
The same brain, eyes and hands show up across very different industries:
- Logistics and warehousing: mobile robots that navigate aisles without fixed tracks, arms that pick and pack mixed items, and cameras that read labels, count stock and measure parcels as they pass.
- Manufacturing: visual inspection that spots defects on a production line, robot arms that load machines or sort parts, and safety systems that slow equipment when a person comes too close.
- Medical technology and healthcare: smart imaging and diagnostic devices that process data at the bedside, lab automation that handles samples, and hospital robots that move supplies. Running AI on the device keeps patient data local. Products used for medical purposes also need to meet TGA requirements, so plan for that early.
- Agriculture: machines that detect weeds, assess crops or guide harvesting in places with poor or no internet coverage.
- Mining, construction and infrastructure: rugged camera systems and autonomous machines for inspection, surveying and work in hazardous areas, where sealed GMSL2 cameras and industrial computers earn their keep.
- Retail and smart buildings: people counting, shelf monitoring and access systems that analyse video on site, rather than streaming it to the cloud.
In each case the pattern is the same: a camera sees, an edge computer decides in real time, and a machine or a person acts on the result.
From prototype to production
Most projects follow the same path:
- Prototype on a developer kit. It's cheap, flexible and quick to start.
- Pilot on a ready-built edge computer. It comes in an enclosure, with the ports and mounting a real site needs.
- Scale with Jetson modules. Design them into your own product once volumes justify it.
Because every Jetson runs the same NVIDIA software, your work carries forward at each step.
Where to start
- Learning or experimenting? An Orin Nano Super Developer Kit and a ZED Mini.
- Automating a task on site? A ready-built edge computer and a ZED X camera.
- Building a robot? A reBot arm, a Jetson and a ZED camera, or see Build a robot.
Not sure which pieces fit together? See Build at the edge and Give machines vision, or book a free consultation.
General guidance only. Figures are approximate and depend on your model, software and settings. Book a consultation before you buy.