NVIDIA is addressing the data scarcity bottleneck in robotics by enabling the creation of training-ready physical AI datasets via its Cosmos platform. Developers can now use as few as two text prompts to generate diverse simulation scenarios, such as rare equipment defects or nighttime incidents, which are traditionally difficult and expensive to capture in the real world.
The system streamlines the workflow of data curation, labeling, and evaluation by integrating previously disconnected tools into a unified generation process. By leveraging world models to produce physically accurate synthetic data, the technology aims to accelerate the development of autonomous systems and industrial robots that require massive amounts of visual and sensory information to operate safely.

