NVIDIA is accelerating innovation in the global robotics industry by unveiling a host of new open models, robotics frameworks, and next-generation AI computing platforms that will usher in the era of physical AI at CES 2026.

NVIDIA unveiled next-generation physical AI models and robots through 'NVIDIA Live at CES 2026 with Founder and CEO Jensen Huang' and 'NVIDIA Showcase' held on the 5th.

Major global companies such as Boston Dynamics, Caterpillar, LG Electronics, and Neura Robotics have contributed to the expansion of the robotics ecosystem by showcasing new robots and autonomous machines based on NVIDIA technology.

NVIDIA CEO Jensen Huang emphasized, “The era of ChatGPT has arrived in robotics,” adding, “Physical AI models that understand the real world, reason, and plan actions will open up new applications across industries.”

NVIDIA supports companies in rapidly advancing the development of next-generation robots through a full stack consisting of Jetson, CUDA, Omniverse, and OpenPhysical AI models.

NVIDIA is launching a new open The model is provided through Hugging Face.

Cosmos Transfer 2.5 and Predict 2.5 are open-world models that support synthetic data generation and simulation-based policy evaluation.

Cosmos Reason 2 is a Vision-Language Reasoning (VLM) model that enables human-level perception, understanding, and action planning.

The Isaac GR00T N1.6 is a VLA model that supports full-body control and advanced context understanding for humanoid robots, and Franca, Neura, and other humanoids are utilizing it to train new robot movements.

Salesforce has reduced robot image analysis time by half by utilizing NVIDIA Blueprint and Cosmos Reason, and medical robot company LEM Surgical is applying NVIDIA Isaac for Healthcare to train autonomous arms for Jetson AGX Thor-based surgical robots.

NVIDIA released OSMO with Isaac Labs-Arena as open source to unify complex robot development environments.

Isaac Lab-Arena is a collaborative framework for large-scale policy evaluation and benchmarking that supports test standardization by integrating with major benchmarks such as Libero and Robocasa.

OSMO is a cloud-native orchestration platform that manages synthetic data generation, model training, and software-in-the-loop testing from a single control center, and it has also been integrated into Microsoft Azure Robotics Accelerator.

NVIDIA is collaborating with HuggingFace to integrate the Isaac and GR00T models into the open-source robotics framework 'LeRobot'.

Through this, 2 million NVIDIA Robotics developers and 13 million HuggingFace AI developers are connected into a single ecosystem.

Rich 2, Rich Mini, etc.Open source humanoid robots are also expected to be integrated with Jetson Thor and DGX Spark to provide customized AI experiences.

NVIDIA unveiled the Jetson T4000 module based on the Blackwell architecture.

It is optimized for autonomous machines and industrial robots by providing AI performance that is four times better than existing ones and 64GB of memory within 70W of power.

In addition, IGX Thor, launching this month, provides functional safety and enterprise-grade AI computing, leading the expansion of physical AI into industrial edge sectors such as aviation and manufacturing.

Neura Robotics, RichTech, AGIBOT, LG Electronics, and others unveiled new Jetson Thor-based humanoid products at CES, and Caterpillar is expanding its cooperation with Nvidia to automate construction and mining equipment.