ASUS Gives Gallop Humanoid a 99.5% Packing Target on Its Own Factory Line

ASUS wants its first humanoid robot, Gallop, to pack 15 motherboard accessories with a 99.5% success rate in no more than 20 seconds per board — and it announced the target on November 13, 2026, alongside a detailed account of how the machine learns.

The task is deliberately hard. Many of the parts sit in thin, clear bags, which makes the workstation a difficult assignment for simple grippers. ASUS picked it to test whether the robot can take on most of a job without a person continuously standing beside it. One caveat matters here: Gallop's figures are targets, not reported achieved results. ASUS applies the same speed-and-success standard to human workers, which sets a clear bar — the robot has to match people, not just perform demos.

The stakes reach beyond one workstation. Electronics manufacturers struggle to staff repetitive, physically demanding stations, and humanoids promise to slot into factories built for people. ASUS is betting its own production lines can prove that promise years before the robot reaches the market.

What can Gallop actually do today?

The complete demonstration, learning and line-testing loop now works. A person shows the task, the robot learns it, and ASUS tests it on the line. That's the foundation; the remaining question is whether the loop holds up under daily factory conditions.

The Humanoid Robotics business unit inside ASUS's Physical AI Solutions Business Group develops the robot, in a program that began in early 2026. The current machine uses a fixed base and two arms for stationary pick-and-place work. ASUS designs the body in-house, along with a touch-sensing dexterous hand — though that hand has not yet been integrated into the robot. Motors and other standard components come from established partners.

ASUS's broader case for the humanoid form is practical: arms and dexterous hands can use existing workstations, tools and trays designed for people, so a factory doesn't need rebuilding around the robot. The company also concedes the limit of that argument — not every task requires a humanoid. The takeaway is that ASUS sees the human shape as a compatibility feature, not a virtue in itself.

How does the robot learn a job?

ASUS starts with open AI models rather than building a giant model internally. The training pipeline runs in four steps:

• Experienced workers, with their consent, wear head-mounted cameras, so their demonstrations show how the job is actually performed.
• Simulation generates variations of those recorded examples.
• An operator wearing a VR headset and gloves guides Gallop through practice with the same parts and cameras used at the real station.
• Recordings made once the robot is deployed feed back into training.

That last step is the flywheel. Every shift the robot works produces fresh data from the real station, which sharpens the next round of training.

ASUS describes its edge as the right data — from real stations, real parts and real quality rules. Its own electronics manufacturing services factories supply the test environment, which most robotics companies must rent. Testing on your own lines means you control the variables and can iterate fast.

What's inside the software stack?

Gallop's software splits into four layers, each with a defined job:

• A planner interprets the scene and splits the job into steps.
• An AI supervisor checks each step and replans after a problem.
• A skills layer handles reaching, grasping and placing.
• A control layer moves the arms and hands precisely and stops them if something is wrong.

The architecture decision that matters most is the interfaces. Fixed interfaces between the layers let ASUS replace AI models without rebuilding the robot. In a field where models improve monthly, that swap-friendly design could keep Gallop from aging as fast as monolithic robots.

The whole system runs on one NVIDIA Jetson AGX Thor computer inside Gallop, using standard robotics software. No cloud dependency in the loop description, and no exotic stack — ASUS is leaning on commodity AI hardware and its own data advantage. The one-line takeaway: the moat isn't the model, it's the factory.

When can customers buy one?

The roadmap is specific. ASUS plans to integrate the in-house dexterous hand, add wheels, and show Gallop to customers at Computex 2027. Customer pilots are also planned for that year. Market availability is targeted for the second half of 2028.

Electronics manufacturing is the first intended market. ASUS expects pick-and-place skills learned in one job to transfer to others, which points to later uses in hospitals and service sites. The company isn't promising those markets yet — it's arguing that a skill learned once shouldn't die on one station.

On jobs, ASUS positions the robot for repetitive, physically demanding stations that are difficult to staff, not as a replacement for workers. Employees who teach Gallop would move into setting up, supervising and improving robots. That framing aligns with how automation actually lands in factories: the teaching role becomes the skilled role.

What has to go right before 2028?

The 2027 pilots carry the real weight. They will test whether learning from skilled workers and station-specific data can sustain factory quality and pace through daily operation — not just through a staged demo. ASUS has made that consistency the threshold for broader use.

Gallop's working training loop is the starting point. The pilots must show that it can produce repeatable results under everyday factory conditions, where parts arrive imperfectly, lighting changes and stations get rushed. Until then, the 99.5% and 20-second figures remain goals, and the humanoid market's hardest problem — consistency, not capability — stays unsolved.