NVIDIA announced the ‘Cosmos World Foundation Model (WFM)’ platform. Because of this we are going to speed up the event of physical artificial intelligence (Physic-AI) systems and expand them to incorporate robots and autonomous vehicles.
NVIDIA CEO Jensen Huang appeared within the opening keynote speech at ‘CES 2025’ held in Las Vegas, USA on the sixth (local time) and unveiled ‘Cosmos’.
“The ‘ChatGPT moment’ for robotics is coming,” said CEO Hwang. “We created Cosmos to democratize physical AI and make general robotics accessible to all developers.”
Cosmos is a development platform consisting of ▲World Model (WFM) ▲Tokenizer ▲Guardrail ▲Video processing pipeline.
Robotics, which must derive responses to commands through actions, is experiencing difficulties in development because of a scarcity of existing data. Cosmos solves this problem by generating physically-based synthetic data.
This can be a method of making ‘physics-based video’ by combining input data comparable to text, images, and video with robot sensor-motion data. Environments comparable to warehouses, factories, and roads will be created with prime quality. After model development and evaluation, it is feasible to simulate various possibilities and situations by combining the ‘omnibus’.
Already, leading robot startups comparable to 1X, Agile, Agility, and Figure AI, in addition to ride-sharing specialist Uber, are collaborating with NVIDIA through Cosmos.
Cosmos plans to be released as open source. Developers can access previews from the NVIDIA API Catalog or download model families and fine-tuning frameworks from the NVIDIA NGC Catalog and Hugging Face.
Subsequently, quite a lot of latest solutions aimed toward the ‘AI agent era’ were also unveiled at the location.
First, ‘AI Blueprint’ helps developers construct and deploy video-image evaluation AI agents.
The non-public AI supercomputer ‘Project Digits’ was also unveiled. This can be a personal AI supercomputer that gives performance at the extent of the Grace Blackwell platform. Users can develop and run inferences from models using their very own desktop systems, then deploy the models to accelerated cloud or data center infrastructure.
As well as, NVIDIA announced that it has established partnerships with Toyota, Aurora, and Continental to develop next-generation autonomous vehicles.
Amongst them, Toyota plans to provide a next-generation vehicle based on ‘DRIVE AGX Orin’.
Reporter Jang Se-min semim99@aitimes.com
