ASUS tests Gallop humanoid on factory packing, with reliability still to prove

Gallop's hand and wrist-mounted camera. Integration of ASUS's own dexterous hand remains on its company-reported development agenda. Image: ASUS. Source: ASUS.

Manufacturing experience becomes training data

ASUS says access to its group factories gives it production parts, experienced operators and quality requirements against which to develop Gallop. Its proposed advantage is task-specific demonstrations and testing conditions, rather than simply collecting more data. ASUS is also the developer and the initial evaluator; its account provides no independent assessment of performance.

According to ASUS, Gallop starts with existing open AI models. Experienced workers, with their consent, wear head cameras to record demonstrations. Simulation produces variations, and an operator wearing a virtual reality headset and gloves guides the robot through the task using the workstation's parts and cameras. Recordings from deployment are intended to feed further training. A report describing the development process attributes this approach and the performance target to ASUS, rather than providing independent test results.

Gallop currently has two arms on a fixed base. ASUS designs the body and is developing a dexterous hand with tactile sensing in-house, while sourcing motors and other standard components from partners. Integration of its own hand and a move to wheels are still on the coming year's agenda.

ASUS describes software that separates planning, supervision, movement skills and physical control. A second AI checks the planner's steps and can revise the plan when something goes wrong. Fixed interfaces are intended to let ASUS exchange AI models without rebuilding the machine. The system runs on a single NVIDIA Jetson AGX Thor onboard computer. That design combines outside computing hardware and models with ASUS's own body, hand and task integration; whether model changes improve factory output will require testing.

Figure's robotic hands and wrists. Its company-reported BMW deployment account identified the forearm as the leading hardware failure point. Image: Figure AI. Source: Figure AI.

Factory use already has a precedent

Humanoid developer Figure AI offers a precedent for both factory operation and the hardware problems it can expose. In a November 19, 2025 deployment account, Figure said its second-generation Figure 02 robot had loaded more than 90,000 parts and logged more than 1,250 hours of runtime during an 11-month deployment at carmaker BMW's Spartanburg plant. Its task was loading sheet-metal parts into a fixture for conventional industrial robots to weld.

Figure described six months of daily runtime within the deployment and said the robot ran 10-hour shifts, Monday through Friday. Those are supplier-reported operating figures for a different task, not a like-for-like benchmark for Gallop. Figure also listed placement success and human interventions as performance measures, but its stated success threshold and zero-intervention goal were targets, not disclosed achieved rates.

The deployment fed into another hardware generation. Although Figure described hardware failures as minimal overall, it identified the forearm as its leading failure point and said it redesigned wrist electronics and cabling for its successor, Figure 03. Figure 02 had returned from BMW as part of its retirement. The account shows why factory access matters beyond demonstrations and AI training: sustained use can reveal component and thermal constraints that require physical redesign.

ASUS's company-reported timeline distinguishes the program's start from planned pilots and targeted market availability. Source: ASUS.

The next proof is sustained operation

ASUS plans a customer demonstration at Computex 2027, customer pilots during 2027 and market availability in the second half of 2028. A brief account sourced to ASUS repeats that timetable; it is a company plan, not a confirmed commercial rollout. Hospitals and service sites are longer-term ambitions based on ASUS's expectation that pick-and-place skills can transfer between jobs.

The company says Gallop is intended for repetitive, physically demanding stations that are hard to staff, with workers moving into robot setup, supervision and improvement. That is its proposed division of labor, rather than a reported employment outcome.

The announcement gives no sustained runtime, human intervention rate, price or operating-cost comparison. Customers will need those measures alongside achieved task accuracy and speed to judge whether Gallop can meet quality requirements while reducing the labor needed for the whole station. The planned pilots must also establish whether the learning process transfers beyond ASUS's own production environment—and whether a humanoid is worth choosing over conventional automation for the job.