Nvidia introduced a new artificial intelligence (AI) model last week designed for training robots through simulation. Named Cosmos-Transfer 1, this large language model (LLM) focuses on AI-driven robotics hardware, often referred to as physical AI. The company has made this model available as open source under a permissive license, allowing interested users to download it from well-known online repositories. Nvidia emphasized that a key benefit of this latest AI model is the enhanced control it offers users over the simulations generated.

The use of simulation for robotics training has recently gained traction, thanks to advancements in generative AI technology. This area of robotics involves hardware that operates using AI Infrastructure as its core intelligence. The training approach prepares the machine's "brain" for various real-world situations, enabling it to perform a broader array of tasks. This represents a significant Technological innovation enhancement over current factory robots, which are typically designed for a single function.

Cosmos-Transfer 1 is part of Nvidia's Cosmos Transfer world foundation models (WFMs), which process structured video inputs such as segmentation maps, depth maps, lidar scans, and more to produce photorealistic video outputs. These outputs serve as a simulation environment for training physical AI.