ASUS’s first humanoid robot is learning to pack motherboard accessories

ASUS is teaching its first humanoid robot, Gallop, to pack motherboard accessories. The company’s November 13 account places the trial at an ASUS group electronics manufacturing plant. It plans to run customer pilots in 2027 and is targeting commercial availability in the second half of 2028. The first test for a factory is whether the robot can pack the complete set that goes with a motherboard at an existing workstation.

In a post that day, ASUS co-CEO S.Y. Hsu said the job involves 15 small parts. Many come in thin, clear bags. A bag can bend as it is lifted, and the item inside can shift. The robot has to find an edge it can grasp and keep control of the package until it reaches the tray. Picking up a single part is one step in a job that has to end with the complete set in the box.

What counts as a successful pack

ASUS is aiming for a 99.5 percent success rate within 20 seconds, the standard it says it uses for workers at that station. The November announcement gave the target. It did not give the robot’s measured pass rate, the number of attempts or the length of a continuous run. Those details would show how consistently it handles the task over repeated cycles.

The U.S. National Institute of Standards and Technology uses task boards to study robot assembly with both rigid pieces, such as screws, and flexible pieces, such as wires and belts. Common test pieces give researchers a way to compare time and handling errors across systems. Its grasping research also looks at how much an object slips when a force is applied. For a packing station, that distinction matters when a flexible bag moves in the robot’s grip after pickup. The NIST studies describe separate test methods and provide no results for Gallop.

What the robot learns from workers

ASUS says experienced workers consent to having their work recorded by a small head-mounted camera. The company uses simulation to vary the examples. An operator wearing a VR headset and gloves also guides the robot through the task with the same parts and cameras used at the station. A worker’s recording shows the order of the job. Guiding the robot supplies an example of how its own arms and hands move through the same task.

A bag that someone can hold flat with a finger might crease or turn in a machine’s grip. The company’s training process gives the robot practice with the objects it will encounter and feeds records from its work back into training. ASUS described starting from open models and adapting them with these examples. It gave no model name or count of the demonstrations used for the trial.

From a fixed base to customer trials

The current machine uses two arms at a fixed station. ASUS says one onboard NVIDIA Jetson AGX Thor computer runs its software for interpreting the scene, planning the next move and checking progress. NVIDIA describes the platform as a robotics computer for processing sensor input and running AI models on the device. ASUS has yet to publish how much energy Gallop uses per pack or how long each decision takes.

ASUS says the software components communicate through fixed interfaces so that it can update the AI models without redesigning the robot. The current job is within reach of the two arms. The company plans to fit a touch-sensing hand it developed itself and to build a wheeled version.

ASUS intends to show Gallop to customers at Computex 2027. The organizer has scheduled the Taipei show for June 1 through June 4. The company’s subsequent milestones are trials with customers in 2027 and a target of market availability in the second half of 2028. Its announcement gave no price or initial shipment volume.