NVIDIA is also expanding its Cosmos open world foundation models to accelerate the development of physical AI systems, including robots and autonomous vehicles that must act safely in the real world.
Because it’s costly and even dangerous to collect varied real-world data, synthetic data generation is becoming the essential solution to train and test these models at scale.
The latest Cosmos Predict 2.5 version unifies Text2World, Image2World, and Video2World into a single lightweight framework capable of generating controllable multi-camera video worlds from text prompts, images, or videos.
Cosmos Transfer 2.5, meanwhile, allows developers to add various weather, lighting, and terrain conditions throughout environments to achieve increased scenario coverage.
Together, these updates are integrated into synthetic data pipelines in NVIDIA Isaac Sim and the Omniverse platform, closing the loop between simulation and real-world deployment.
Top robotics firms, including Skild AI, Serve Robotics, and Zipline, already utilize Cosmos and Isaac Sim to generate massive synthetic datasets that enhance reliability, reduce costs, and shorten training times.
Synthetic data is also improving industrial inspection, logistics efficiency, mining operations, and robotics handling tasks by enabling photorealistic, physics-accurate training without safety risks.
With growing demand for physical AI, synthetic worlds are becoming a key enabler for the next generation of autonomous machines.
You may also want to check out some of our other recent updates.
Wanna know what’s trending online every day? Subscribe to Vavoza Insider to access the latest business and marketing insights, news, and trends daily with unmatched speed and conciseness! 🗞️
