FANUC Corp. last week said it has strengthened the integration between its ROBOGUIDE simulation software and the NVIDIA Isaac Sim open robotic simulation reference framework. The company said it will enable intuitive operation within a “virtual factory” and help realize accurate, integrated digital twins.

FANUC is using imitation learning and the NVIDIA GR00T foundation model to enable one of its robots to fold T-shirts. The company noted that its system uses the NVIDIA Jetson Thor platform.

Editor’s note: FANUC is a 2026 RBR50 Robotics Innovation Award winner for helping Wilson Bohannan increase U.S.-based production. See the RBR50 honorees on the show floor and at the ticketed RBR50 gala dinner at the Robotics Summit & Expo next week. Register now to attend.

---

---

‘Virtual factory’ runs on Isaac Sim and ROBOGUIDE

At the International Robot Exhibition (IREX) last December in Tokyo, FANUC demonstrated technology that imports robot motion simulations created in ROBOGUIDE into NVIDIA Isaac Sim. It reproduced precise trajectories and cycle times in a virtual environment using the same control algorithms as the actual robot.

With the enhanced integration, FANUC said the two systems can now deliver a more practical and efficient simulation environment for robot testing and virtual commissioning.

The first mode of integration places NVIDIA Isaac Sim at the forefront, with ROBOGUIDE operating in the background to ensure accurate robot behavior in the virtual space. ROBOGUIDE is tightly integrated with Isaac Sim, with continuous direct communication.

“In this new environment, users can intuitively operate robots in Isaac Sim in real time from virtual or physical teach pendants connected to ROBOGUIDE—just as if they were controlling an actual robot,” said FANUC. “Users can perform jogging operations, teach robot programs, execute the programs, and verify results directly within Isaac Sim. This enables intuitive and efficient pre-installation studies and process design for robots within GPU accelerated physically accurate sensor and environment simulations.”

The companies asserted that the NVIDIA Isaac Lab open robot learning framework and NVIDIA Omniverse libraries enable high-precision simulations of tasks that were traditionally difficult to reproduce. Examples include handling flexible components like cables or performing insertion and assembly operations.

Robots operating in Isaac Sim maintain identical trajectories and cycle times to real machines through integration with ROBOGUIDE, removing the “sim-to-real gap,” claimed the partners. This combined environment also supports reinforcement and imitation learning, accelerating the evaluation and deployment of physical AI systems, said FANUC and NVIDIA.

The second integration mode places ROBOGUIDE at the forefront, while the NVIDIA PhysX physics engine enables advanced simulation in the background. ROBOGUIDE can now use PhysX for accurate simulation of complex tasks such as bin picking, which were previously difficult to simulate.

Randomly piled parts can be realistically simulated using physics-based modeling, while ROBOGUIDE’s 3D vision system identifies part positions for pick-and-place operations. Real-world scenarios—such as determining when a robot cannot extract a specific part and selecting an alternative—can be replicated virtually, said FANUC.

“This advancement allows users to complete feasibility studies for bin-picking systems in a virtual environment, eliminating the need for extensive trial-and-error using actual parts,” asserted the company. “Combined with NVIDIA PhysX, ROBOGUIDE significantly improves the efficiency of designing and deploying bin-picking systems that previously required considerable time and expertise on site.”

GR00T N allows FANUC cobots to fold T-shirts

At an open house this month, FANUC demonstrated a dual-arm system using two CRX collaborative robots to fold flexible objects — in this case, T-shirts — by imitation learning. It used cameras for visual recognition and the NVIDIA Isaac GR00T N open robot foundation model.

Folding flexible objects like T-shirts requires continuous adaptation of robot motion to changing shapes, noted FANUC. Such folding is difficult to achieve with conventional playback teaching or vision-based robot path compensation, it said.

The company will demonstrate an operator performing the folding task using CRX robots in real time. The system will learn from these examples through imitation learning. The dual-arm CRX will be trained to replicate the task and acquire the necessary techniques to complete the folding process.

Traditionally, imitation-learned robot motions tended to appear segmented and jerky. By combining FANUC’s advanced motion control technology with the NVIDIA GR00T N model, the system will have smooth, continuous movement, said the companies.

NVIDIA Jetson Thor platform

At IREX, FANUC also introduced an “AI robot that avoids human” based on its open platform. The system has been upgraded with NVIDIA’s latest robotic computer, Jetson Thor.

By replacing NVIDIA Jetson AGX Orin module with Jetson T5000 module, the system’s AI compute has improved by more than 7.5 times, asserted the companies. Jetson Thor allows the robot to avoids human movement more effectively, said FANUC.

The company’s new product exhibition this month also features the operation of robots in a virtual space, high-precision simulations, physics simulations using NVIDIA PhysX, demos of dual-arm robots trained through imitation learning, and cobots featuring the latest NVIDIA edge computing platform.

The post FANUC strengthens robot integration with NVIDIA Isaac Sim appeared first on The Robot Report.