On-Premises Industrial Training Center

When an industrial client evaluates bringing humanoid robots into their operation, the hard part of the conversation is almost never the robot.

It's three other questions: Where does my plant's data end up? Who owns the skills the robot learns? And can this run inside my operation, without depending on a connection to the cloud?

They're the right questions. And they're exactly the ones our On-Premise Training Center is built to answer.

The idea is simple: the client operates their own Training Center inside their plant. The entire lifecycle of a robot's skills data capture, training, inference, and deployment runs inside the client's perimeter. The data never leaves their infrastructure. And the client retains full intellectual property and total governance over everything their robots learn.

The node that makes it possible: NVIDIA IGX Thor

Running the full cycle at the plant takes serious compute at the edge. That's where NVIDIA IGX Thor comes in.

IGX Thor is NVIDIA's industrial-grade edge computing platform, built on Blackwell architecture and designed for physical AI in regulated environments manufacturing, energy, logistics, healthcare. This isn't consumer hardware: it's an enterprise-grade platform with functional safety built in and long-term support.

In client terms, here's what matters:

Datacenter-class power, on your floor. IGX Thor has enough capacity to run, in parallel and on a single node, the models that give the robot its intelligence: action models (Isaac GR00T), perception and reasoning models (Cosmos) all running locally.

Instant response. Local inference and teleoperation with latency under 100 ms, with no dependency on the internet. The robot responds instantly because the intelligence lives next to the robot, not on a remote server.

Works offline. The entire operation can run with no external connection. The cloud is optional, not a requirement. If your plant is in a low-connectivity area, it doesn't matter.

Functional safety out of the box. The hardware is designed for functional safety certification and aligned with industrial standards, with an architecture built for environments where safety is non-negotiable.

Industrial time horizon. 10 years of guaranteed hardware and NVIDIA software-stack support. It's an infrastructure asset with an industrial amortization horizon not technology that goes obsolete in a couple of years.

Data governance: full sovereignty, inside your four walls

This is where most conversations about industrial AI get uncomfortable and where we made a deliberate design choice.

The data captured when you train a robot on your operation is some of the most sensitive data you have: it's a detailed map of how your plant works. That data should never be forced out to the cloud as a condition of using the technology.

With the On-Premise Training Center, it doesn't leave. All operational and training data stays at the plant. Full sovereignty: you decide where your data lives and who ever gets to touch it.

And this lines up with what your operation already has to comply with. The architecture is designed hand-in-hand with a cybersecurity roadmap that accounts for standards like IEC 62443 and ISO 27001  for clients in automotive, energy, and oil & gas, where a facility map is a security matter, not just a business one.

Intellectual property: the hardware, the data, and the models are yours

This is the part clients don't always see coming, so we'll be direct: when you train robots on your operation, you create something that didn't exist before skills learned from your environment, impossible to replicate in any other plant.

In this model, that belongs to you entirely. The client owns the hardware, their data, and their models. Humandroid provides the platform Robots ID along with implementation and operation under a subscription model. But the intelligence built inside your plant is yours, and it stays with you.

Two consequences follow from that.

It's unique to you. Nobody else operates your plant, so nobody else can produce the same skills. What's learned inside your walls is a competitive advantage, not something that leaks into a shared model everyone else benefits from.

It appreciates. The more your robots work, the more real-world data they gather, and the sharper and more reliable their skills become. You're not buying a frozen capability you're building one that compounds over time, and that becomes an asset of your company.

How it fits into Robots ID

Our product, Robots ID, is a 7-layer operating system for training and deploying humanoid robots. IGX Thor is the on-site deployment node: the physical piece where the operational layers run.

Here's what that enables:

• Flexible training. Models can be trained in the cloud or, in fully on-premise mode, directly on the client's own infrastructure.

• The same software everywhere. IGX Thor runs the same NVIDIA stack we use to train. What works in development works in production, with nothing to rewrite.

• Real-time operation. The node processes the robot's sensors, runs the AI policies, and enables teleoperation with minimal latency.

• Multi-manufacturer. A single Training Center can operate robots from different manufacturers Unitree, UBTech, and others thanks to our hardware abstraction layer.

• Local fleet management. Monitoring and orchestration of the robots run inside the plant.

The robot on the floor: Cruzr Y1

If IGX Thor is the intelligence, the Cruzr Y1 is the body that puts it to work — and it's built for industrial pilots that don't stay pilots.

Ubtech Cruzr Y1

The Y1 is a wheeled humanoid with dual 7-DOF arms and a 10 kg payload, ready for real deployment in handling, palletizing, loading/unloading, and sorting. It's designed so you can move directly from pilot to large-scale industrial rollout, without swapping platforms halfway through.

It's also built to actually stay on the floor. With 4+ hours of operation at full load, auto-charging, and a quick-swap battery, the Y1 covers 24/7 shifts without downtime.

And it's a Robots ID native. Operated through our platform, the Y1 can be trained on new industrial tasks and anchor an on-premise Training Center — following a structured roadmap of skills that scale from first training to full production deployment.

Where it goes to work:

• Handling

• Depalletizing

• Loading & unloading

• Sorting

What this means for you

Bringing humanoids into an industrial operation isn't, at its core, a hardware question. The robot is the easy part.

The hard questions are the three from the start, and our answer is consistent across all of them:

Your data stays in your plant. Your skills and the models that hold them are yours. And it all runs on-premise, on NVIDIA IGX Thor, under Robots ID, operated by us as a service.

That's what an On-Premise Training Center is really for. It's not where you go to see a robot.

It's where your robots learn to do your work inside your walls, with your data, and where the value they create stays with you. 💚