ByNeetika Walter

AI and RoboticsMay 15, 2026 05:54 PM EST

NVIDIA-powered FANUC robots can train and validate tasks before real-world deployment.FANUC

Japanese firm FANUC and NVIDIA have expanded their robotics partnership to build factory robots that behave the same way in virtual simulations as they do in the real world, a step aimed at speeding up industrial automation and reducing costly on-site testing.

The integration combines NVIDIA Isaac Sim with FANUC’s ROBOGUIDE simulation software, allowing engineers to test, train, and validate robotic systems inside physically accurate digital environments before deployment.

The companies say the upgraded system creates tightly connected digital twins where virtual robots reproduce the same trajectories and cycle times as real machines using identical control algorithms. FANUC first demonstrated the technology at the International Robot Exhibition in Tokyo last year, but the new version deepens communication between the two systems.

In the first setup, NVIDIA Isaac Sim acts as the main virtual environment while ROBOGUIDE runs in the background to ensure robot behavior remains synchronized with real-world motion. Engineers can operate robots inside Isaac Sim using physical or virtual teach pendants connected to ROBOGUIDE, enabling real-time jogging, programming, execution, and verification inside a simulated factory floor.

The setup also uses NVIDIA Omniverse libraries and Isaac Lab to simulate difficult robotics tasks such as cable handling, insertion operations, and assembly work that are traditionally challenging to recreate accurately in virtual environments.

Closing simulation gaps

One of the biggest challenges in robotics is the “sim-to-real” gap, where robots trained in simulation fail to behave the same way in real-world conditions due to differences in physics, timing, or environmental interactions.

FANUC says its integration with ROBOGUIDE keeps robot trajectories and cycle times identical between simulation and physical deployment, helping remove those inconsistencies. The environment also supports reinforcement learning and imitation learning for AI-powered robotic systems.

The second mode of integration places ROBOGUIDE at the forefront while NVIDIA PhysX handles physics simulation in the background. This enables more realistic testing of industrial tasks such as bin picking, where robots must identify and retrieve randomly piled parts.

Using physics-based modeling, the system can simulate scenarios in which robots fail to extract one object and autonomously select another, reducing the need for repeated real-world testing with physical parts.

The companies say this virtual feasibility testing could significantly cut deployment time for complex automation systems that previously required extensive on-site tuning.

FANUC also upgraded its AI-powered human-avoidance robot using NVIDIA’s Jetson Thor platform, increasing compute performance by more than 7.5 times compared to the earlier Jetson AGX Orin-based system.

Robots learn cloth folding

Alongside the digital twin technology, FANUC demonstrated a dual-arm robotic system that learns to fold T-shirts using NVIDIA’s Isaac GR00T N robot foundation model.

The setup uses two CRX collaborative robots trained through imitation learning, where a human operator first performs the folding task, and the robots learn from those demonstrations.

Flexible objects such as clothing are particularly difficult for robots because the shape continuously changes during handling. FANUC says the robots generate motion in real time while visually tracking the object using cameras.

The company added that combining its motion control system with NVIDIA’s GR00T N model produces smoother movements than traditional imitation-learned robot systems, which often appear segmented or jerky.