Nvidia launches GR00T 2 and Isaac Lab 3.0 at CoRL (www.therobotreport.com)

NVIDIA has unveiled GR00T 2, an advanced open frontier world action model, and Isaac Lab 3.0, an upgraded open-source robot learning framework, during the recent Conference on Robot Learning (CoRL) in Austin, Texas. GR00T 2, trained on 100,000 hours of diverse robot and human data, showcases impressive capabilities, achieving 71% success in various tasks with minimal demonstrations and enabling skill transfers across different robot models. This model ranks among the top performers in vision-language-action benchmarks, providing researchers powerful tools for developing and annotating robot-training datasets. Meanwhile, Isaac Lab 3.0 allows rapid setup of training environments with improved physics modeling, supporting the simulation of complex robotic interactions and achieving significantly higher training throughput compared to its predecessor. The significance of these releases lies in their potential to accelerate robotic learning and performance in real-world applications. Isaac Lab 3.0 integrates seamlessly with the newly introduced Newton physics engine, which enhances the accuracy of simulations, crucial for transferring learned skills to real robots. Furthermore, the introduction of open benchmarks like DexBench and VABAR offers a framework for rigorous evaluation of robotic dexterity and adaptability, pushing the boundaries of what robots can achieve in various industrial tasks. As robotics companies adopt and implement these advanced technologies, it is expected to foster innovation and improve operational efficiency across sectors such as logistics, manufacturing, and hospitality.