ASUS revealed a two-armed humanoid robot called Gallop on November 13, 2026, and the most striking detail is where it is already working: on the shop floor of an ASUS group electronics manufacturing services factory, packing 15 types of motherboard accessories, many of them small, thin objects sealed in soft plastic bags. The performance target is 99.5% task success within 20 seconds, matching the speed of a trained human worker. If Gallop hits that number in full production, ASUS will have become the first consumer-electronics company to close the loop, designing and manufacturing the hardware that assembles its own hardware.
Inside the Factory Test: 15 Accessory Types and a 20-Second Clock
The workstation ASUS chose for Gallop's first real-world assignment was not engineered for a robot. It was designed for a human. According to Interesting Engineering, ASUS engineers explicitly wanted Gallop to demonstrate that a humanoid can step into an existing production environment without custom fixtures or redesigned tooling. The 15 accessory types include items that frustrate conventional automation: soft plastic bags, thin connectors, and components that deform under grip. These are precisely the objects that robotic arms with rigid fingers fail on and that keep expensive manual labor irreplaceable in electronics factories.
Gallop's current form is a fixed-base platform with two arms and dexterous hands developed entirely in-house, with tactile sensing embedded in the fingers. Motor systems come from industry partners, though ASUS has not disclosed which suppliers. Inside the robot sits a single NVIDIA Jetson AGX Thor compute module, the same platform NVIDIA designed for robot inference workloads running multiple neural networks simultaneously. ASUS plans to release a wheeled version within the coming year, extending the robot's reach beyond a single workstation to mobile multi-station coverage.
The training methodology combines three inputs that reinforce each other. Workers who consent wear head-mounted cameras to record demonstrations, giving the robot a first-person view of how a skilled human approaches each part type. Simulation generates variations of object placement, orientation, and packaging state to broaden the training distribution beyond what any single human demonstrator would produce. A VR headset with gloves allows a remote operator to guide the robot's hands in real time, providing correction data when it encounters a situation the simulation did not cover. Deployment data from live factory runs feeds back into the training loop, so Gallop is designed to improve while it works, not just before it ships.
The Four-Layer Architecture That Keeps AI Models Swappable
The engineering choice that may matter most for Gallop's long-term competitiveness is not the hardware but the software architecture. ASUS built Gallop around four distinct layers: planning, supervision, skills, and low-level control. Each layer communicates through fixed interfaces. That design means the AI model running any given layer can be replaced without redesigning the system around it. As foundation models for robot manipulation improve, with NVIDIA's GR00T platform, Physical Intelligence's pi0, and others competing, ASUS can swap in a better model at the planning layer without touching the control stack.
This modular approach is a deliberate hedge against the pace of AI development. Factory deployments commit equipment for years. A robot locked to a specific model vintage becomes obsolete as the field advances. By separating the AI components from the mechanical and control layers behind stable interfaces, ASUS gives itself, and its future customers, a software upgrade path that does not require hardware replacement. The architecture also allows different models to run different layers simultaneously, so a lightweight model can handle real-time control at low latency while a larger model handles high-level task planning at a slower cadence. That split mirrors how modern autonomous vehicle stacks separate perception from planning from actuation.
The in-house dexterous hand with tactile sensing is the hardware bet that matters most. Touch sensing lets the robot detect when a part is about to slip or when it has grasped a bag incorrectly before the error propagates. Most commercial robot hands on the market lack tactile feedback, which forces them to rely entirely on vision, and vision fails when objects are deformable or transparent. ASUS says the production version of its hand is coming within the next year, which is a faster timeline than many dedicated robotics companies have managed for a comparable capability.
Why Electronics Manufacturing Is the Right First Target
Not every factory is an equal proving ground for a new humanoid. ASUS's choice of electronics assembly is shrewd for reasons beyond the company's own familiarity with the domain. Electronics factories have several structural properties that make them suited to humanoid trials. First, the product mix changes frequently, requiring workers to learn new assembly procedures every few weeks. This is exactly the kind of adaptability that rules out conventional fixed automation and makes general-purpose robots attractive. Second, the parts are small, expensive, and fragile, so quality control demands fine motor control rather than raw strength or speed. Third, the labor cost base in Taiwan and Southeast Asian electronics hubs is rising steadily, narrowing the cost advantage that made manual assembly economical for decades.
ASUS also has an advantage no outside robotics company can replicate: a captive test environment. The EMS factory where Gallop is running is part of the ASUS group. The team can observe failures, retrain on live production data, and iterate without the commercial pressure of a paying customer watching the error rate. ASUS Press showed at Computex 2026 that the company is building an AI robot portfolio including the Kairo service robot and the Maestro companion robot. Gallop is the manufacturing arm of that portfolio, and the EMS factory is the controlled laboratory where the technology has to earn its deployment at scale before ASUS sells access to anyone else.
Compare this with the approach of other humanoid makers. Figure AI's Figure 02 robots are deployed at a BMW factory in Spartanburg, South Carolina, with high public visibility and low tolerance for downtime. Agility Robotics' Digit works in Amazon fulfillment centers, in an environment specifically designed around robot ergonomics. ASUS's factory floor was not redesigned, which makes the success bar harder and the proof more transferable to customers who cannot afford to retrofit their facilities. TweakTown reported ASUS chairman Jonney Shih's direct ambition to compete with Tesla Optimus in the humanoid market, and the captive factory deployment is the evidence-base behind that claim.
The Taiwan angle deserves more attention than the product announcement has received. ASUS is not an isolated actor. The company sits at the center of an electronics supply chain that includes Foxconn, Pegatron, Wistron, Inventec, and Compal, all of whom run assembly lines facing the same labor dynamics. If Gallop works inside ASUS's factories, the pressure on those contract manufacturers to adopt similar systems will be immediate. Taiwan's electronics OEM ecosystem has historically moved together: when one player upgrades manufacturing technology, the others face competitive pressure to match it or risk losing contracts to the upgrader. ASUS building Gallop in-house rather than licensing from an outside robotics company signals that it intends to sell or license the platform across that ecosystem, which would transform Gallop from an internal cost-reduction tool into an infrastructure product with a much larger addressable market than motherboard accessory packing alone.
The 2028 Commercial Launch and the Competitive Clock Running Against ASUS
The timeline ASUS has laid out is tight. The program began in early 2026. The company will show Gallop publicly at Computex 2027. Customer pilots are planned for 2027. Commercial availability is targeted for the second half of 2028. That is roughly two and a half years from program launch to general availability, which is aggressive for a robot that has to meet 99.5% task success in an unstructured production environment.
The 2028 window puts ASUS into direct competition with Agility Robotics, which has been shipping Digit to enterprise customers since 2024 and will have several additional years of real customer data by then. Boston Dynamics' Atlas electric humanoid is also moving toward industrial deployment under Hyundai's ownership. Humanoid Hub tracks deployment milestones across the sector, and the pace of real-world factory integrations has been accelerating through 2026, meaning ASUS is entering a market that will be more crowded and more demanding by the time its commercial product is ready.
The wheeled version coming within the next year matters for the commercial economics more than for the technology proof. A fixed-base robot handles a single workstation. A mobile robot can move between workstations, covering multiple tasks in one shift and recovering autonomously from a failed grasp by repositioning rather than stopping for human intervention. The economics of a mobile humanoid handling ten tasks are dramatically better than a fixed arm handling one, and the mobile form factor is what most factory customers will want to buy. ASUS building the wheeled version now, while the fixed version is still in testing, suggests the team is thinking ahead to the commercial product rather than optimizing for a demonstration.
The Counter-Case for a Consumer Brand Entering Robotics
ASUS is a world-class consumer electronics and PC hardware company. It is not a robotics company, and the bear case for Gallop is specific: building a production-grade humanoid robot requires years of hardened systems engineering, field-failure databases, and actuator-level expertise that ASUS does not yet have. Critics argue that ASUS will find, as many hardware entrants before it have, that the gap between a working prototype and a 99.5%-reliable production system is measured in years, not months. Figure AI, Agility Robotics, and Boston Dynamics each have years of fielded robot data and teams built entirely around bipedal locomotion and dexterous manipulation. ASUS is starting that clock in 2026, and those companies started it years earlier.
The 99.5% target itself is worth scrutinizing. That figure means one failure in every 200 attempts. At 20 seconds per task, a robot completing full eight-hour shifts handles roughly 1,440 tasks per day. At 99.5% success, it fails 7 times per shift. Each failure requires human intervention to recover the part, reset the workstation, and potentially inspect the product for damage. Whether 7 interventions per shift is acceptable depends entirely on the alternative, and in a high-mix electronics factory with scarce skilled labor, it may well be. But the number illustrates how demanding the standard is for a system that is still in internal testing. The gap between current field performance, which ASUS has not disclosed publicly, and the 99.5% target is likely substantial, and that gap is what the next two years of internal iteration must close before pilots begin.
What to Watch Before Computex 2027
Three signals will tell you whether Gallop is on track before the public demonstration. First, watch for ASUS to announce the number of factory stations it has deployed internally. Any number above five suggests the system is reliable enough to run with minimal human supervision. A number at one or two workstations suggests the team is still managing the robot manually between tasks, which would be far from the commercial product's reliability threshold.
Second, watch the dexterous hand announcement. ASUS says it will release its in-house hand within the coming year. The quality of that hand, and whether it can handle the full 15 accessory types without fixture modification, is the core technical question. If ASUS licenses or acquires an external hand instead of shipping its own, that is a signal the in-house development hit a barrier that required a faster solution. Third, watch the customer pilot announcements in 2027. The company says pilots are planned for that year, but it has not named any pilot partners. Electronics contract manufacturers like Foxconn, Pegatron, and Wistron are the natural first customers, since they also run ASUS-brand product lines and share the same assembly challenges. A pilot with one of those three companies would validate that Gallop generalizes beyond ASUS's own factory. A pilot only with ASUS subsidiaries would suggest the technology is not yet ready for external production environments. NVIDIA's Jetson AGX Thor compute partnership is also worth watching: if ASUS deepens the relationship toward NVIDIA's Isaac robot platform for training, it gains access to synthetic data and foundation model tools that could materially accelerate the gap-closing between current performance and the 99.5% commercial threshold.
The company that assembles your GPU just built the robot that may one day assemble it for you.
Key Takeaways
- 99.5% task success within 20 seconds is the performance target for Gallop packing 15 types of motherboard accessories at an ASUS group EMS factory, matching trained human worker speed
- NVIDIA Jetson AGX Thor powers Gallop's inference, with a four-layer AI architecture (planning, supervision, skills, control) designed so individual AI models can be swapped without redesigning the system
- Program began early 2026, with Computex 2027 demo, 2027 customer pilots, and commercial availability targeted for H2 2028, putting ASUS on a two-and-a-half-year launch timeline from inception
- Fixed base with two arms in current testing form; a wheeled mobile version is planned within the coming year, the configuration that makes multi-station factory economics viable for paying customers
- ASUS's captive EMS factory provides a testing advantage no outside humanoid maker can replicate: live production data without commercial pressure, allowing an aggressive iteration pace before 2027 pilots begin
Questions Worth Asking
- If ASUS achieves 99.5% task success for motherboard accessories, which electronics assembly tasks would still require human workers, and what does that imply for the residual labor market in Taiwan electronics manufacturing?
- What is the production cost of Gallop relative to the annual labor cost it replaces, and at what failure-rate threshold does the economics break down for a customer without ASUS's internal tolerance for downtime?
- Does ASUS's entry into humanoid robotics signal that Taiwan's OEM ecosystem will vertically integrate into automation, reshaping the competitive map before North American and European robot makers realize it has happened?
