NVIDIA has recently launched Cosmos, a cutting edge platform that is equally as designed to advance the progression of physical AI technologies such as robots and self-driving cars. Announced at CES, this platform presents seven world foundation models (WFMs) which are neural networks capable of predicting and generating physics-aware realistic avatars of digital spaces. These WFMs have been built for a sole purpose, to help the developers to train, test and optimize their AI systems to the best possible precision possible.
Today, WFMs are gradually turning into indispensable tools in physical AI construction, like large language models for prosaic language processing. These WfMs employ data in the format of text, images, video, or motion to build representations of virtual worlds with reference to real physical connections. To prompt the discovery of new possibilities and to help speed up the process of applying physical AI applications, these models are provided under open license by NVIDIA regardless of whether the party is a researcher, developer, or business. This brings ease in the access of WFMs and also reduces hurdles that are associated with the incorporation of these WFMs in different projects.
Cosmos has several model categories including Nano for instant deployment in edge locations, Super for leading benchmark outcomes, and Ultra for customized premium qualities. These models have been popularized by big firms such as 1X, Agility Robotics, XPENG, Uber, and Waabi for enhancing the prospects of their robotics and self-driving vehicles.
Cosmos has been privileged to work with NVIDIA, which has equipped it with advanced tools to help developers, including tokenizers, safety features, as well as a framework for model flexibility. Thus, by strengthening the models with NVIDIA NeMo framework, there is the ability to fine-tune the model with your specific dataset to ensure versatility in your specific application. In addition, Cosmos models are compatible with NVIDIA Omniverse, and this makes it easy to generate synthetic video data for training the AI.
The Cosmos WFMs have acquired essential expertise on a vast 9,000 trillion tokens, acquired from over 20 million hours of data-Intensive preparation, giving the Cosmos WFMs an explosive advantage in the creation of physics-based synthetic data, simulating hypothetical environments, and reinforcement learning. These advanced capabilities are valuable particularly for industries like automobile and robotics where conducting actual experiments is costly and tedious. One of them is used by Waabi where Cosmos is used to filter video data for its AI operated simulator called Waabi World. Likewise, Hillbot has adopted this type of technology where it has designed sophisticated 3D maps that help to reduce the time of training robots significantly.
Cosmos also has a major focus on the responsible use of artificial intelligence advancement in technology. This can be evidenced by Cosmos Guardrails which are used to limit the effects of adverse inputs and encourage safe use of the AI created content. Furthermore, the software has an in-built watermarking system to enhance the level of transparency since some sequences can be produced by the AI system. All these elements are aligned with NVIDIA’s belief system on responsible AI, the company’s principles of safety, privacy, and ethical conduct.
The efficiency of the platform is supported by the use of NVIDIA DGX Cloud, a relatively recent infrastructure with GPUs that is capable of processing 20 million hours of video data in as little as 14 days. This rapid and complex processing is a paradigm shift to the traditional CPU structures, which would take several years to accomplish the same task. Moreover, the Cosmos tokenizers offer higher compression rates and faster rates, which lead to lower computational costs for training as well as predicted results.
With Cosmos, NVIDIA has prepared the conditions for a revolutionary period in the world of physical artificial intelligence. It offers solid and easy-to-use tools to developers to build and model virtual worlds on this site to speed up the development cycle and bring novel features in robotics and auto systems in the market more efficiently.
Source: https://blogs.nvidia.com/blog/cosmos-world-foundation-models/
Latest Stories:
AI to Reshape Asia-Pacific Economies, Risks Widening Inequality
Unveils Samsung Vision AI and Next-Generation Screen Innovations at CES 2025
MIT Researchers Develop AI System to Revolutionize Research Hypotheses
What is your reaction?
