A Foundational Software Platform for Scaling Hardware-Agnostic Control of Digital Twins and Physical Systems

Originally published by READY Robotics CEO, Benjamin Gibbs.

We built a foundational software platform that enables a completely hardware-agnostic sim-to-real-to-sim feedback loop, which enables future technologies like AI to more easily scale across any robot or industrial automation system in a way previously not possible.

“The unique architecture of ForgeOS allows for the sim-to-real workflow we need for this production line because it bridges Nvidia Isaac Sim’s simulation capabilities with the control of our physical system. This means we can program in simulation, transfer that program, and then use ForgeOS to capture live production data and iteratively improve our processes.” Kazahiro Suzuki, Toyota group manager

What did we build?

Last week, we announced a project with NVIDIA and Toyota Motor Corporation that combines high-fidelity simulation capabilities of NVIDIA Omniverse with ForgeOS, the highly-extensible industrial OS for robotics and automation built by our company READY Robotics .

This is the first ever truly hardware-agnostic implementation of a sim-to-real-to-sim feedback loop that enables physical robots and industrial automation equipment, as well as their digital twins, to be controlled using the same REST APIs. Toyota will be using this powerful, combined software architecture of ForgeOS and Omniverse to automate its aluminum hot forging production lines. Traditionally, programming robotic systems for forging involves working with hot metal parts, posing significant safety challenges. The ForgeOS + Omniverse integration enables seamless programming in a simulated environment, eliminating risks associated with hot parts.

In the article from Dean Takahashi of VentureBeat that announced this project, Kazuhiro Suzuki, group manager of Toyota’s raw materials development division, had this say: “The unique architecture of ForgeOS allows for the sim-to-real workflow we need for this production line because it bridges Nvidia Isaac Sim’s simulation capabilities with the control of our physical system. This means we can program in simulation, transfer that program, and then use ForgeOS to capture live production data and iteratively improve our processes.”

How does AI fit into this?

We're not claiming to have built any new AI here, but what we have built is a software platform that provides AI agents and models with a bridge to interact with and control robots from almost any major brand, as well as thousands upon thousands of industrial automation components.

I have long believed that the biggest bottleneck in massively scaling robotics (and more broadly, industrial automation) is the fact that all this very powerful hardware, very sophisticated hardware lacks a modern software layer. ROS tried to be this. I love the vision of ROS, but ROS is really just a collection of middleware services.

What is truly needed is an integrated platform built from the ground up with a modern software architecture.

So that's what we built with ForgeOS.

ForgeOS runs on the edge, using an off-the-shelf industrial PC, connecting to and controlling physical automation hardware in manufacturing facilities.

ForgeOS also runs in the cloud, connecting to and controlling virtual automation hardware with high-fidelity physics simulated in NVIDIA Omniverse.

As the physical hardware executes an automation program, ForgeOS feeds all the data of the system back into the digital twin in real-time.

AI uses that data to learn about the characteristics and performance of the physical system.

AI can then write new automation programs to optimize performance, test them in the Omniverse simulation, and then before push that exact program to the real-world system via the ForgeOS APIs. No need to deal with antiquated software stacks from legacy robot and automation OEMs (who make amazing hardware by the way!)

We've been working on ForgeOS for a long time, and its deployed in manufacturing facilities around the world. We know it works in the real-world.

Now thanks to the ongoing collaboration with NVIDIA that started in 2022, ForgeOS can now control high-fidelity simulated systems as well.

We'll be sharing more details and examples of how this combined software architecture works in the coming days and weeks, so please stay tuned!