NVIDIA improves physical AI models with synthetic data generation through updates to NVIDIA Cosmos and Omniverse. Cosmos Predict 2.5 unifies models for consistent video world generation, while Cosmos Transfer 2.5 offers style transfer for data variation. These technologies are utilized by companies like Skild AI and Serve Robotics to enhance robotics policies and autonomous deliveries. Serve Robotics recently used its robots to deliver NVIDIA DGX Spark personal AI supercomputers. Zipline also participated in the DGX Spark drop for its drone delivery systems. Lightwheel and data scientist Santiago Villa are using synthetic data to improve robotics and mining operations. FS Studio partnered with a logistics leader to enhance package detection accuracy using Omniverse libraries. Robots for Humanity and Omniverse Ambassador Scott Dempsey utilize synthetic data for simulation environments and cable detection applications. Stay updated on OpenUSD, Cosmos, and synthetic data for physical AI through NVIDIA Omniverse resources and communities. Join discussions on the Alliance for OpenUSD forum and AOUSD website.

Read more at NVIDIA: Scaling Physical AI With Synthetic Data