NVIDIA swaps traditional simulation for accurate physics in robotics

NVIDIA unveils new robotics technologies for simulating delicate tasks. The company is launching GR00T 2 and Isaac Lab 3.0, with Newton integration for accurate physics. The technology could help companies train robots with greater realism.

NVIDIA Cosmos is a platform that offers world models and tools for developing physical AI. The technology addresses two challenges: adapting models to sensors, tasks and data, and integrating those models into applications reliably. The company is expanding its robotics partnerships in South Korea with LG Electronics and Hyundai Motor Group.

LG Electronics begins partial operations at its robot training factory, while LG Innotek expands collaboration on sensing technologies with NVIDIA. NVIDIA is introducing new benchmarks, such as DexBench, to evaluate dexterity and industrial adaptability. Open tools make it easier to collect data and train robots. The collaboration with partners such as LG Electronics, LG Innotek and Hyundai Motor Group, especially in South Korea, aims to develop more adaptable and efficient robotics solutions.

NVIDIA is launching GR00T 2, a world action model that can perform 40 of 50 delicate tasks in simulation. This means robots trained with this technology are better able to handle precise movements, such as assembling electronic components or handling fragile objects. Integration with Newton, an advanced physics system, lets robots better understand the force and direction needed for each action, reducing errors and increasing efficiency.

The technology also connects to Isaac Lab 3.0, which offers tools for developing physical AI, making training more flexible and adaptable. Companies that use industrial robots can benefit from more realistic simulations, which speed up learning and reduce the cost of physical testing. NVIDIA Cosmos, in turn, helps integrate these models into practical applications, ensuring that robots work reliably in real-world environments.

What's new is the ability to simulate complex interactions with the environment, something that was previously limited to closed, controlled environments. NVIDIA is also expanding its partnerships in South Korea, where companies such as LG and Hyundai are already integrating its technologies into robot training factories. The technology allows robots to adapt to varied scenarios, reducing the need for constant reprogramming. This is crucial for companies that operate dynamic production lines, where flexibility is a differentiator. Isaac Lab 3.0 integrates Newton for accurate physics, enabling more realistic and robust simulations in dynamic environments.

With this, robots learn to interact with objects in more natural ways, such as handling fragile parts or adjusting posture in real time. This reduces development time and increases the reliability of tests in real-world environments. With these tools, companies can train robots more efficiently, cutting costs and increasing productivity. NVIDIA is expanding its robotics partnerships in South Korea with LG Electronics and Hyundai Motor Group, which have already begun operations at robot training factories. The collaboration could accelerate the deployment of solutions in markets with high demand for automation.

With information from blogs.nvidia.com, developer.nvidia.com and digitimes.com.

Text written with AI from the sources listed and checked by the newsroom's automated rules. How we work

• #hardware
• #software