American tech giant Nvidia has launched an advanced and innovative artificial intelligence (AI) model that is designed to improve the training of AI-powered robotic systems through simulation.
The large language model (LLM), named Cosmos-Transfer1, is all set to give developers powerful control over simulation environments, making it an important and valuable model resource for those working on robotic training.
The model has been released as open-source, keeping in mind the interested developers and researchers on popular platforms like GitHub and Hugging Face. The firm Cosmos-Transfer1 is the latest addition to its Transfer World Foundation Models (WFMs).
Additionally, simulation-based training is becoming increasingly popular in the robotics sector. Further, it is also developing hardware that utilizes AI as its core processing unit.
As per the reports, the firm’s AI model uses the structured video inputs, such as segmentation maps, depth maps, lidar scans, and more, to create high-quality, photorealistic video outputs.
These outputs can train the AI-powered robots, enabling them to learn from diverse simulated environments. According to a paper published by Nvidia in the arXiv journal, Cosmos-Transfer1 offers superior customization compared to previous models.
Nvidia stated that, “It allows the weight of different conditional inputs to vary based on spatial location, enabling developers to create highly controllable simulation environments.”
The AI model is a diffusion-based design equipped with seven billion parameters and is optimized for video denoising in the latent space. Its control branch takes text and video inputs and will create photorealistic output videos. It also supports four types of control input videos, like Canny edge, blurred RGB, segmentation mask, and depth map.
Cosmos-Transfer1 has been fully tested on Nvidia’s Blackwell and Hopper series chipsets, with inference conducted on the Linux operating system. Its design permits real-time world generation, delivering a more efficient and diverse training experience for AI systems.
Nvidia has made the Cosmos-Transfer1 AI model functional under the Nvidia Open Model License Agreement, which permits both academic and commercial use of it. Developers and researchers can download the model from Nvidia’s GitHub and Hugging Face listings.
Also Read: “India Should Give Befitting Reply to OpenAI’s Sam Altman”: MP Raghav Chadha
The Crypto Times – Read More