At the GTC Berlin conference, NVIDIA announced a significant expansion of its Isaac open robotics development platform, introducing a new humanoid robot reference platform, runtime framework, simulation assets, and datasets, along with updates to robot learning tools. These enhancements cover the entire workflow from training data generation and simulation training to physical robot deployment. In the hardware domain, NVIDIA partnered with Agile Robots to launch the humanoid robot reference platform Agile ONE SN, set to debut in early 2027. The platform comes pre-configured with Isaac GR00T and is powered by the Jetson Thor computing platform, designed to accelerate robot R&D and significantly reduce initial setup time. Standing at 1.4 meters tall, NVIDIA reports the Agile ONE SN weighs 40 kilograms, while Reuters reports approximately 34 kilograms, with the reason for the discrepancy unclear. The robot features 31 degrees of freedom, a single-arm payload capacity of 3 kilograms, and is equipped with two five-fingered dexterous hands, supporting 1kHz force control and tactile sensing. Its Jetson Thor computing platform delivers up to 2070 TFLOPS of FP4 compute performance and 128GB of unified memory, along with various perception sensors. Future plans include integrating NVIDIA's Halos full-stack safety system. Several renowned overseas research institutions have expressed strong interest in the platform, with NVIDIA and Agile Robots planning to collaboratively produce up to 2,000 units by 2027, primarily targeting university and corporate research lab customers. On the software front, NVIDIA officially released the open-source robot learning framework Isaac Lab 3.0, which supports multi-GPU training workflows and adapts to various computing environments. Simultaneously, the open-source runtime framework Isaac OS, built on Holoscan, was introduced. It connects various robot-related components, supports migration from simulation to physical robots, and offers deterministic real-time operation capabilities. Additionally, NVIDIA released three open-source reference workflows for the DGX Station, covering GR00T model optimization, dexterous manipulation skill training, and world model post-training synthetic data generation. For data support, the SimReady Central Library provides over 2,000 simulation robot and factory assets, while the Human 1M dataset includes multi-view videos and 3D hand-tracking data of various human activities. Both data resources are available via the Hugging Face platform, with the full version of Human 1M exclusively accessible to Cosmos Coalition partners. Notably, several industrial robotics companies have already adopted NVIDIA's robot development tools and related platform technologies.