According to Nvidia, they’d just launched the world’s best robotics platform. This is a big deal because it is the first time we’ve seen artificial intelligence embedded into such physical machines. The company’s latest offerings include the Cosmos world models, specifically Cosmos Transfer 2.5 and Cosmos Predict 2.5, designed for synthetic data generation and robot policy evaluation in simulations. To support this new direction, Nvidia has released Cosmos Reason 2. This VLM focused on reasoning with multimodal language is supercharging AI systems’ ability to see, read, and interact with the world around them.
At a recent Accelerate Conference from our friends at Nvidia, they announced something that should really get you excited. They outlined the burgeoning importance of robotics as AI transforms from cloud-based systems to machines that can learn and reason in real-world conditions. This transition is supported by advancements in cheaper sensors, enhanced simulation capabilities, and AI models that can effectively generalize across various tasks.
With the announced new models, Nvidia Physical AI has developed a strong ecosystem. These models include open foundation models that unlock the potential for robots that can reason, plan, and adapt to new tasks and environments. This is a foundational approach to developing unified standards across the robotics industry. Further, it will assure interoperability and increase throughput across multiple robotic technologies.
Perhaps one of the most significant innovations at play is the Isaac GR00T N1.6 model. This next-generation vision language action (VLA) model is particularly adept for humanoid robots. With Cosmos Reason as its centralized processing unit, humanoid robots powered by GR00T can help them avoid obstacles, run, jump, and even perform backflips to achieve whole-body control. It gives them the ability to both pick up and push objects all at once, which extends their usefulness across more applications.
In addition to Nvidia’s latest announcements on Arm’s server business, this year also brought Jetson T4000 graphics card powered by Blackwell, which joins Thor family of products. This new addition is a reflection of Nvidia’s dedication to delivering powerful hardware solutions that power its powerful AI models and growing robotics efforts.
Nvidia is indeed doing the outreach by releasing its models on Hugging Face. This platform has driven the boom in robotics-related downloads, with Nvidia’s tools for the platform topping the chart. This trend is emblematic of the increasing appetite for robotics in AI’s deep end.
Nvidia’s Isaac Lab-Arena serves as a welcome one-stop resource hub. It offers challenging task scenarios for the community to experiment with, training tools, and established benchmarks (such as Libero, RoboCasa, and RoboTwin). This environment fuels the creativity of academic researchers, independent developers, and large companies alike, providing strong incentives for innovative collaboration and competition in the industry.
Rebecca is really excited about all the things that Nvidia is doing to make the future of robotics and AI bright. To confirm this news story or for more information, please contact her by email or through Signal or other encrypted messaging apps.