National Robotics Week: Latest Physical AI Research, Breakthroughs, and Resources

The latest robotics technology unveiled at NVIDIA GTC focused on combining simulation, robot learning, and edge computing into a single cloud-to-robot workflow. Key announcements include Isaac GR00T open models, Cosmos world models, the open-source physics engine Newton 1.0, and the GA of Isaac Sim 6.0, Isaac Lab 3.0, and Omniverse NuRec.

This combination allows robots to understand natural language instructions, perform complex multi-step tasks, and learn faster using large-scale synthetic data. The workflow for validating behavior in near-realistic environments before real-world deployment has been further strengthened.

Medical and industrial field applications were also introduced. PeritasAI is pursuing operating room automation and multi-agent collaboration using NVIDIA Isaac for Healthcare and the Rheo blueprint, while Doosan Robotics implemented adaptive palletizing with Cosmos Reason that reasons about box contents, damage, and handling methods.

Research projects also stood out. NemoClaw converts natural language commands into Python scripts to control Nova Carter within Isaac Sim, and OceanSim uses GPU-accelerated underwater simulation to rapidly generate sonar and synthetic images. RoboLab is a benchmark for comparing and evaluating the performance of generalist robot policies through high-fidelity simulation, and it will also be reflected in the future Isaac Lab-Arena roadmap.

The trend toward open-source and edge robotics is also accelerating. OpenClaw aims for fully local execution using Jetson Thor, Nemotron, and vLLM, while Skyentific's bipedal robot project and the University of Maryland's household humanoid research demonstrate the potential of simulation-first development and Jetson-based deployment.