NVIDIA Showcases GR00T 2 and Isaac Lab 3.0 at CoRL for Advanced Robot Training

NVIDIA has unveiled groundbreaking advancements in robotics at the 2026 Conference on Robot Learning (CoRL) in Austin, Texas. The company introduced GR00T 2, a world action model, and Isaac Lab 3.0, an open-source robot-learning framework, to enhance the precision and scalability of robot training. These innovations aim to bridge the gap between laboratory demonstrations and real-world deployment by enabling robots to perform complex, dexterous tasks with greater autonomy and adaptability.

GR00T 2: A Leap in World Action Models

GR00T 2, built on NVIDIA Cosmos, is designed to train robots in dexterous manipulation using diverse datasets, including 100,000 hours of open-source human egocentric data. Key features of GR00T 2 include:

• Out-of-the-box performance: Successfully completed 40 of 50 evaluated tasks without task-specific demonstrations.
• Skill transfer: Enabled robots to learn from human demonstrations and adapt skills across different robot bodies.
• Generalization with minimal data: Achieved a 71% success rate across 50 tasks with one demonstration per task, rising to 95% with five demonstrations.
• Agent control: Provided tools for AI agents to sequence skills and recover from errors.

GR00T 2 ranks among the top three models across benchmarks such as RoboCasa365, RoboLab, RoboDojo, and MolmoSpaces. It is being tested by Nexuni for robotic laundry handling, Foxconn for factory automation, and Siemens for industrial applications. MicroAGI has also utilized GR00T 2 for Ethernet cable swapping tasks using a YAM robotic arm.

Isaac Lab 3.0: Scaling Robot Training in Simulation

Isaac Lab 3.0 provides a robust platform for training robots in simulated environments before deploying them in the real world. The latest version introduces:

• Enhanced physics integration: Supports deformable objects and complex articulated systems using the Newton physics engine.
• Improved training throughput: Achieved up to 2.5x the training speed of its predecessor, Isaac Lab 2.3, in vision-based dexterous manipulation benchmarks.
• Flexible simulation tools: Allows researchers to switch between NVIDIA PhysX and Newton solvers, select renderers for perception-based training, and apply reinforcement post-training libraries to foundation models like GR00T.

Skild AI announced the STEEL solver, a rigid-body physics engine built on Newton, to further accelerate robot training with Isaac Lab. Flexion demonstrated how Isaac Lab and Newton support its Reflect model, combining reasoning and whole-body manipulation.

Collaborative Efforts and Broader Impact

NVIDIA is working with partners such as Ambi Robotics, Bosch Research, Siemens, and UC Berkeley to bring these advancements to the robotics community. The company also introduced new workflows, including Video to Data, which converts human video into simulated robot demonstrations, and EgoHand, a 3D hand motion reconstruction model with over 40,000 downloads on Hugging Face within two weeks of launch.

By combining simulation, world action models, and agentic robotics workflows, NVIDIA’s latest technologies aim to accelerate the development of robots capable of performing intricate tasks in dynamic environments, paving the way for broader industrial adoption.