NVIDIA's Isaac platform provides an end-to-end cloud-to-robot workflow for building generalist-specialist robots. The stack covers synthetic data generation via Omniverse NuRec and 3D Gaussian splatting, teleoperation data collection with Isaac Teleop, policy training in Isaac Lab 3.0 with parallel simulation environments, and deployment on Jetson edge hardware. Key components include the open GR00T VLA model for robot reasoning, the Newton physics engine for realistic simulation, the Physical AI Data Factory Blueprint for scalable data pipelines, and SOMA-X for standardizing skeleton and motion representations across platforms. Safety tooling (NVIDIA Halos), open datasets, and educational resources round out the offering announced at GTC.
Sort: