Pune, India | January 06, 2026

NVIDIA has released advanced physical AI models designed to help developers create smarter, more capable robots for real-world tasks. The announcement occurred at CES 2026 in Las Vegas, emphasizing the company’s strategy to push robotics forward. Furthermore, these models allow machines to perceive environments accurately, reason logically, and act efficiently in diverse settings.

The new models include Cosmos Transfer 2.5, Cosmos Predict 2.5, Cosmos Reason 2, and Isaac GR00T N1.6. As a result, robotics engineers can simulate real-world scenarios, generate synthetic training data, and optimize robot behavior much faster than traditional methods. These models are available on platforms such as Hugging Face, supporting collaboration among developers worldwide.

Physical AI models empower robots to perform complex tasks autonomously, analyze environments intelligently, and execute precise actions. Consequently, development teams can focus on innovation rather than repetitive programming, significantly improving productivity and reducing time-to-market.

Global robotics leaders have already embraced NVIDIA technology. Boston Dynamics, Caterpillar, Franka Robotics, Humanoid, LG Electronics, and NEURA Robotics demonstrated robots powered by physical AI models. Therefore, these machines now achieve higher levels of autonomy, efficiency, and accuracy in both industrial and domestic applications.

NVIDIA CEO Jensen Huang highlighted this release as a transformative moment for robotics, describing it as a “ChatGPT moment.” Moreover, he explained that combining Jetson processors, CUDA software, Omniverse simulation tools, and open physical AI models provides a full-stack solution for building advanced autonomous robots.

The Jetson T4000 module, powered by Blackwell architecture, offers up to four times higher AI compute performance. As a result, it can serve as the central processor for industrial automation robots, autonomous surgical systems, and smart home-assistance devices.

Alongside hardware, NVIDIA introduced frameworks such as Isaac Lab Arena and OSMO to unify robot training workflows across cloud and edge environments. Isaac Lab Arena enables large-scale benchmarking and evaluation of robot behaviors in simulation. Consequently, robots can reliably perform tasks before deployment in real-world conditions. OSMO orchestrates synthetic data generation, model training, and automated testing, streamlining development workflows for robotics teams.

Opening physical AI models to the public has drawn significant attention from the developer community. NVIDIA’s collaboration with Hugging Face integrates the models with the popular LeRobot open-source framework. In addition, this allows millions of engineers and hobbyists to access powerful AI tools, accelerating global innovation in autonomous machines.

Data scarcity, a key challenge in robotics, is addressed through physical AI models. Cosmos models simulate diverse scenarios, therefore providing expansive datasets without the costs and risks of physical testing. This capability allows developers to create more resilient, adaptive, and intelligent robots.

Industry adoption extends beyond traditional robotics. Salesforce uses physical AI models with real-time video analysis to reduce operational incidents, while LEM Surgical trains autonomous surgical arms. Meanwhile, applications in logistics, manufacturing, and healthcare illustrate the broad versatility of these models.

The announcement reflects AI’s evolution from digital-only tasks to embodied, real-world applications. Consequently, robots are becoming smarter, more adaptable, and capable of performing tasks autonomously in dynamic environments.

NVIDIA’s open-stack strategy emphasizes empowering developers, accelerating AI-driven robotics adoption, and enabling scalable, intelligent machines. Moreover, the physical AI model ecosystem encourages collaboration, experimentation, and rapid innovation globally.

By combining robust hardware, open-source software, and collaborative frameworks, NVIDIA and partners are advancing robotics development. Thus, physical AI models are transitioning from experimental tools into practical, deployable systems across industries worldwide.