Simulating worlds: Agile Robots early access to NVIDIA Cosmos 3

Agile Robots is pushing the boundaries of Physical AI – the intelligent combination of AI and Robotics. And we are doing so in collaboration with the strongest AI names in the world. As NVIDIA launches NVIDIA Cosmos 3, we are excited to be amongst those key partners given early access to put it to work, and to be able to make it available for our customers.

Cosmos 3 brings together synthetic world generation, vision-reasoning and action simulation in one unified world foundation model. Our AI team has been testing and exploring what this new platform can do, particularly in the area of simulation.

A partnership built over years

Agile Robots has already worked closely with NVIDIA for years, leveraging their platforms and compute solutions to train, test, and deploy AI solutions for our customers. We use NVIDIA Isaac Lab, an open framework for robot learning, to design, simulate, and test robotic behaviours in highly realistic virtual environments. We deploy NVIDIA Jetson for onboard AI inference. And as an anchor customer of the European Industrial AI Cloud – a joint initiative by Deutsche Telekom and NVIDIA – testing and integrating Cosmos 3 seamlessly into our platform is the latest chapter in this close collaboration. This allows customers and developers to leverage its capabilities directly.

Testing Cosmos 3 – and making it available for our customers

Our team has been testing Cosmos 2.5 and now Cosmos 3 across our robotics portfolio, including with Thor single- and dual-arm and FR3 dual-arm robots. That has focused on simulation – exploring how Cosmos 3 can serve as a powerful foundation for realistic, scalable virtual environments that we and our customers can use to test, validate and optimise robotic workflows. We’ve been testing and deploying Cosmos 3 as a neural simulator: a learned world model that can generate realistic environments and scenarios for testing and validating robot behaviours. We can evaluate different robot policies directly inside these neural simulation environments, stress-testing edge cases, prototyping new capabilities, and accelerating experimentation. For customers this can help to reduce dependency on real-world trials and speed up deployment cycles.

Why this matters

Achieving generalised robot intelligence is no easy task. It requires not just capable and dexterous hardware, but vast amounts of data, robust models, and simulation infrastructure capable of training robots to navigate edge cases, adapt to new environments, and perform reliably at scale.

That is the goal Agile Robots is built to meet. We bring together robotics hardware, versatile software, state of the art AI, and deep industrial expertise to deliver physical AI for industry. With experience from more than 20,000 robotic solutions deployed worldwide, and strong partnerships, we have the tools and the expertise to keep pushing the frontier. For us, and for our customers.

Looking ahead

Our early testing on simulation shows the strong capabilities of Cosmos 3. We will enable seamless integration and usage of Cosmos 3 for our customers – giving them new options for simulating, testing, training, and ultimately deploying the next generation of Physical AI. Agile Robots is also exploring NVIDIA Video Augmentation Skill built on top of Cosmos for automating data curation, generation, and evaluation.

We are excited to see where it leads: congratulations to our friends and partners at NVIDIA on the launch of Cosmos 3! 

If you would like to find out more, make sure to read NVIDIA's press releases on Physical AI and Cosmos 3.

NVIDIA Launches Cosmos 3, the Open Frontier Foundation Model for Physical AI: nvidianews.nvidia.com/news/nvidia-launches-cosmos-3-the-open-frontier-foundation-model-for-physical-ai

NVIDIA Releases Major Collection of Open Source Agent Tools and Skills for Physical AI: https://nvidianews.nvidia.com/news/nvidia-releases-major-collection-of-open-source-agent-tools-and-skills-for-physical-ai?ncid=so-link-296657&linkId=100000424463380