ASUS Gallop humanoid robots train on motherboard packing with a 99.5% factory target

A PC and server maker just entered the humanoid robot race from its own factory floor, not from a trade-show stage. On November 13, Dr. Albert Chang published ASUS's first long-form look at Gallop, the company's first humanoid, and Co-CEO S.Y. Hsu introduced it on CNBC the same day. Gallop is not touring viral tricks. It is learning to pack motherboard accessories at an ASUS group EMS plant, with a stated goal of 99.5% success inside a 20-second cycle, the same bar ASUS says it holds people to. For buyers, sellers, and operators on Buy My Robots, that framing matters more than another biped reel. When an OEM owns the line, the parts, and the quality rules, resale questions shift toward logged pass rates, training data provenance, and whether software stacks survive when the robot leaves the fence.

What ASUS disclosed on November 13

Gallop comes from the Humanoid Robotics business unit inside ASUS's Physical AI Solutions Business Group. Today it works from a fixed base with two arms. ASUS describes its strength as pick-and-place at a single station, with a wheeled version next on the roadmap.

The first workstation packs 15 small parts per board, many in thin clear bags. ASUS says the job takes about 20 seconds per motherboard and that the team chose the station because simple grippers struggle on small, soft, and irregular pieces. The company reports the full teaching loop is running: a person demonstrates the task, the robot learns it, and the unit is tested on the line. ASUS has not published current success rate numbers against the 99.5% target, only the target itself.

Compute is a single NVIDIA Jetson AGX Thor inside the robot, on top of what ASUS calls standard robotics software. The company designs the body and an in-house dexterous hand with touch sensing, and sources motors and other commodity components from partners. Customer pilots are planned for 2027, market availability is targeted for the second half of 2028, and ASUS plans a customer showcase at Computex 2027 after it adds the full hand integration and wheels.

How Gallop learns, layer by layer

ASUS's post reads like a factory training manual, not a model card. The company says it starts from leading open models rather than training a giant foundation model from scratch. Workers wear a small head camera with consent so Gallop can watch how the job is actually done. Simulation generates variations of each demonstration. An operator in VR headset and gloves guides the robot on the same parts and cameras the production station uses. After deployment, what the robot records feeds back into training.

Under the hood, ASUS describes four cooperating layers. A planner interprets the scene and breaks work into steps. A second supervising model checks each step and re-plans when something fails. A skills layer handles reach, grasp, and place. A control layer drives the arms and can stop instantly if something goes wrong. Interfaces between layers are fixed so newer AI models can be swapped in without rebuilding the hardware. That modularity is a marketplace footnote: a used Gallop may be only as valuable as the policy bundle and supervisor revision that ship with it.

Why an EMS giant building its own humanoid changes resale math

Electronics contract manufacturers already appear throughout the humanoid archive as deployers of other people's bipeds. Foxconn's Houston AI-server talk tied named NVIDIA stacks to unnamed OEM bodies. Hyundai's supply chain stories put Boston Dynamics Atlas near automotive lines. ASUS is taking a different path. It is the OEM, the EMS operator, and the robot developer in one corporate stack.

That vertical slice creates a specific secondary-market profile if Gallop ever leaves captive plants. Listings would need more than leg footage and Jetson branding. Buyers should ask which station policy is licensed, whether worker-demo video or VR teleop traces are part of the training record, and if the in-house hand firmware is separable from ASUS service accounts. ASUS says Gallop is aimed at repetitive, hard-to-staff stations, not a wholesale labor replacement. Until pilots begin in 2027, treat November's disclosure as architecture and KPI intent, not as a catalog SKU with a price.

The fixed-base form factor also sets expectations. Gallop today is closer to a dual-arm humanoid workcell than a walking logistic carrier. Operators comparing it to listings for Unitree G1 research units, Agility Digit tote runners, or Figure 03 production bots should match morphology to task before they match press-release adjacency.

Checklist before you price an ASUS-origin humanoid

When Gallop or derivative units eventually surface outside ASUS affiliates, diligence should mirror what the company itself says it judges:

Documented cycle time and success rate at the packed station, not only the 99.5% aspiration.
Which open-model baselines, planner builds, and supervisor checkpoints are installed, and whether updates require ASUS cloud keys.
Hand revision, tactile calibration dates, and spare finger or skin inventories if the dexterous end effector is in use.
Whether training relied on worker head cameras or VR demonstration, and what privacy or labor agreements cover that data off-site.
Fixed-base versus wheeled configuration, and whether the listing includes the station fixtures ASUS used in EMS tests.

None of those fields appear in today's release. That is normal for a day-one factory narrative. It is also why marketplace premiums should wait for pilot contracts with measurable output.

What to watch through Computex 2027

Short term, watch whether ASUS publishes actual pass rates on the 20-second accessory pack, including miss counts per shift. Watch the wheeled Gallop variant and whether mobility changes the same KPI or opens new stations. Watch if NVIDIA or ASUS issue joint benchmarks on Jetson Thor throughput for multi-layer planner stacks.

Longer term, watch hospital and service-site language against electronics manufacturing claims. Pick-and-place skills may transfer, but regulatory, cleaning, and patient-handling stacks are different resale animals than motherboard bags. For now, Gallop is a credible signal that another Tier-one hardware brand would rather prove humanoids on its own line than rent someone else's floor.