Nvidia is pushing the boundaries of artificial intelligence (AI) with the launch of Cosmos, a groundbreaking family of foundational AI models designed to enhance how robots and machines interact with the physical world. Announced during CEO Jensen Huang's keynote at CES 2025 in Las Vegas, Cosmos is a significant leap forward, focusing on teaching AI systems to understand and interact with real-world environments, a key step in advancing robotics.
Unlike traditional language models like ChatGPT or Bard, which are trained on vast amounts of text to generate written content, Cosmos is specifically designed to work in the visual and physical realms. Huang explained that the AI models were trained on 20 million hours of real footage, capturing human actions such as walking, hand movements, and object manipulation. This focus on physical actions, rather than abstract concepts, allows Cosmos to teach machines how to recognize and respond to dynamic real-world scenarios, such as navigating spaces or reacting to accidents in industrial settings. "It’s not about creating art," Huang emphasized, "It’s about teaching AI to understand the physical world."
During his CES presentation, Huang showcased Cosmos in action by simulating a warehouse environment where boxes fell off shelves—demonstrating how robots can learn to handle real-world situations with remarkable accuracy. This type of realistic video generation is crucial for training robots to respond to unpredictable events, such as accidents at work or navigating cluttered environments. Furthermore, businesses can tailor Cosmos for their specific needs by incorporating proprietary data into the model, fine-tuning it to their unique requirements.
The potential applications of Cosmos span across numerous industries. Humanoid robot startups like Agility Robotics and Figure AI, as well as self-driving car companies such as Uber, Wasabi, and Wayve, are already harnessing this technology. These companies aim to push the boundaries of what robots can achieve, from complex navigation tasks in uncertain environments to performing delicate operations in real-world settings. Cosmos is playing a pivotal role in driving these innovations, enabling robots to interact with the world in ways that were previously unimaginable.
In addition to Cosmos, Nvidia unveiled an upgrade to its Isaac robot simulation platform, designed to further enhance the capabilities of robotics. The new feature allows robots to learn new tasks with unparalleled efficiency. For instance, developers can provide just a few examples of a desired action, such as grabbing an object, and Isaac will generate synthetic training data, enabling the robot to master the skill quickly and accurately. This improvement allows for more rapid and scalable training of robots, making them better equipped to handle various tasks in dynamic environments.
The keynote also featured life-sized digital images of humanoid robots from industry giants like Tesla, Boston Dynamics, and Agility Robotics. This visual display underscored Nvidia’s ambition to play a central role in shaping the future of robotics, as well as its broader vision of integrating AI into the very fabric of daily life. "With Cosmos and Isaac, we’re giving developers the tools to create robots that can truly understand and operate in our world," Huang stated.
Through these innovations, Nvidia is positioning itself as a leader in AI and robotics. The fusion of AI with robotics has the potential to revolutionize numerous industries, from logistics and manufacturing to autonomous driving. Cosmos is paving the way for machines to not only think like humans but also move and interact with the physical world in a way that brings us one step closer to a future where robots can seamlessly coexist with humans in everyday environments. With Cosmos and Isaac, Nvidia is laying the groundwork for an AI-driven future where machines are not just tools but active participants in shaping our world.
