New NVIDIA Cosmos 3 Open Models for Interactive World Simulation
NVIDIA has released four open-weight checkpoints in its Cosmos 3 family for interactive world simulation and driving data generation. All are post-trained variants of Cosmos 3 Nano.
What the article says
The four models have 16B trainable parameters, though the model cards list 15B. Cosmos3-Nano-Transfer-Auto generates synchronized multi-view driving video across up to 11 cameras, including fisheye, with joint RGB-LiDAR generation and adjustable weather, lighting and time of day. Cosmos3-Nano-Sim-Auto is a streaming variant that produces camera and LiDAR chunks in a few denoising steps for interactive rollouts. Cosmos3-Nano-Sim-Transfer converts simulator controls (depth, segmentation, edges, blur) and a text prompt into photorealistic video. Cosmos3-Nano-Sim-Bimanual predicts video from robot or camera actions for two-arm YAM and AgiBot embodiments.
Post-training recipes, a four-stage training pipeline and agent skills are available on GitHub. The models use the OpenMDW-1.1 license, which permits commercial and non-commercial use. The Bimanual setup requires Linux, an NVIDIA GPU, Docker and NVIDIA Container Toolkit.
Read the official article AnnuIA profile
Summary written by AnnuIA in its own words from the official page of Hugging Face. The content and rights belong to the publisher: please read the original article.