Industrial-grade AI Humanoid Robot and Wafer Handling Robot
Also featuring a live demonstration of a remote-controlled dexterous hand
At the Foxconn Technology Day HHTD25, the highlight of the “Smart Manufacturing” exhibition was the robots and AI-related applications. This event showcased the industrial-grade AI humanoid robot, wafer handling robot, remote-controlled dexterous hand, and service humanoid robot for the first time. Additionally, it shared insights on the operation of AI robots in precision assembly workstations based on digital twins, demonstrating how AI helps robots “train”.
Foxconn’s deep collaboration with NVIDIA has developed an industrial-grade AI humanoid robot, which includes wheeled robots that feature high mobility efficiency, long endurance, and simple maintenance, making them particularly suitable for long-term operations in flat production line environments; another type is the legged robot, capable of climbing stairs and overcoming obstacles, with movements closer to those of humans, suitable for complex factory terrains.
The live demonstration at Technology Day showcased real production line requirements, including dual-hand interface panel handling to expand the operational range with a tilted torso; dual-hand collaboration for transporting larger objects, utilizing a mobile chassis and torso lift to enhance the operational range; even executing high-precision screw fastening tasks, demonstrating the dexterous hand’s control capabilities, even when screw holes are obstructed by pipes, the robot can use one hand to move the pipe aside and use the other hand to insert the screw. These tasks previously relied heavily on manual labor, which not only involved high repetition and labor intensity but also risked precision due to fatigue. Now, with robot assistance, operations can be stable 24/7, improving efficiency and significantly reducing long-term operational costs.
Behind the robot’s operation, a complete technical layout must first exist within the NVIDIA Omniverse libraries ecosystem! We utilize the NVIDIA Isaac Sim open-source robot simulation framework to test and train robots in a highly realistic virtual environment, greatly shortening development time. Specifically, the robot’s “brain” is based on the Isaac GR00T N1 open architecture visual-language-action (VLA) model—data is collected through GR00T-Teleop in virtual space, human demonstration actions are automatically converted into robot trajectories using GR00T-Mimic, and GR00T-Gen randomly generates variables such as lighting and background to enhance environmental adaptability; all training is completed in parallel at scale in the Isaac Lab, far exceeding traditional methods.
Smart Manufacturing Highlight – Wafer Handling Robot SHR-F20
Another highlight of smart manufacturing is the wafer handling robot SHR-F20. This robot is specifically designed to handle wafer carriers and loading/unloading tasks in semiconductor factories, which previously required manual labor; now, the SHR-F20 robot can complete these tasks fully automatically, saving time and reducing the risk of errors. The robot’s biggest feature is its autonomous navigation and safety design.
Combining laser SLAM and QR code technology, it can accurately locate itself within the factory, moving flexibly without being affected by complex spatial environments. Equipped with diagonal dual radars, side radars, and front visual sensors, it can achieve 360-degree environmental detection, avoiding collisions with people or equipment, ensuring high safety.
The design of the SHR-F20 robotic arm is also very flexible, allowing for the addition of 2D or 3D vision, as well as interchangeable grippers and storage modules. This means it is not just a fixed-function robot but can be flexibly adjusted according to different application needs, whether handling various carriers, precise alignment, or tackling more complex operational scenarios.
Another “Remote-Controlled Dexterous Hand” serves as an example of human-robot interaction, allowing direct experience of the dexterous hand’s control feel through a designed remote control mechanism at the HHTD event. In front of the camera, computer vision is used to analyze hand posture, converting hand movements in real-time into signals for the dexterous hand. This year, Foxconn has completed the development of its first dexterous hand, which features a joint direct drive design, has 4 fingers, 16 degrees of freedom, weighs only 1 kilogram, yet can carry up to 5 kilograms.
Smart Manufacturing Highlight – “AI Robot Precision Assembly Workstation”
Another major highlight of smart manufacturing is the “AI Robot Precision Assembly Workstation” based on digital twins, where the robots in the workstation act like automated masters that can think and train themselves, responsible for high-precision assembly of parts. Through digital twin technology, they first “train” in the virtual world. AI conducts reinforcement learning in the virtual environment, continuously trying different assembly methods to discover the smoothest actions and shortest assembly times. Once training is complete, the results are directly synchronized back to the real production line, allowing the robots to operate accurately upon deployment, requiring almost no manual adjustments.
At the same time, Foxconn showcased a factory-wide visualization operation system built on NVIDIA Omniverse libraries, covering three key scenarios: overall construction of new factories, automated production line design, and data center energy management, creating an integrated intelligent decision-making hub from planning, simulation to operational maintenance. Its core value lies in the ability to validate and continuously optimize the entire manufacturing system in a virtual environment, whether for new construction or factory renovation.
From factory layout, production line configuration to energy strategies, all decisions can be iteratively simulated in the digital twin space, significantly reducing the need for rework after physical construction, greatly shortening factory construction cycles, and lowering trial-and-error costs.
In the new generation manufacturing operation system, real-time production line data can be synchronized to the virtual environment, achieving dynamic visualization, remote monitoring, and feedback control. Management personnel, regardless of their location, can grasp equipment status, production capacity performance, and anomaly alerts in real-time, and simulate response strategies in the virtual space, quickly deploying them to the actual production line.
This is also Foxconn’s core competitive advantage in the future expansion of global smart manufacturing layouts.