Asus targets 99.5% success for Gallop humanoid
Asus is building Gallop, a humanoid robot meant to prove it can work a real factory station for a full shift to the same quality standard as a person. Its first job is packing motherboard accessories at an ASUS group electronics manufacturing (EMS) plant, with a target of 99.5% success within 20 seconds. S.Y. Hsu, ASUS Co-CEO, recently introduced Gallop on CNBC. Asus now details what it is building, how the robot learns, and what is still being worked on.
Why a humanoid, and why Asus
Asus argues that AI can now see, decide, and move a machine, and that the parts are largely ready. What is still missing is proof that a humanoid robot can work at a real workstation, every day. Videos of robots dancing, sorting blocks, and folding laundry do not answer a factory manager's question: can it do this station, for a full shift, to my quality standard?
Most factories are built around people, with stations, tools, and trays sized for human hands. Traditional automation is excellent for fixed, high-volume tasks but struggles with work that changes from one production run to the next. A robot with two arms and dexterous hands can use the same stations and tools, so the factory does not have to be rebuilt around it. Asus says not every task needs a humanoid, and it chooses the ones that make sense.
Asus has spent decades integrating hardware and software, from motherboards to PCs to servers, and already builds the kind of computing a robot depends on. It also has its own EMS factories to test in, on real parts, under real quality rules, which most robotics companies have to rent.
Gallop: How ASUS Is — Asus
The first job
The packing station has 15 small parts, many in thin, clear bags, and takes about 20 seconds per board. Asus chose it because it is hard: small, thin, and soft parts are exactly what a simple gripper can't handle. It also tests the rule Asus judges itself by: a robot has to take on most of a job, because if a person still has to stand beside it the whole time, nobody gains anything. The full loop now works: a person shows the task, the robot learns it, and it is tested on the line.
How Gallop learns
Teaching Gallop works much like training a new colleague:
- General knowledge: Asus doesn't build its own giant AI model. It starts from leading open models that already understand how hands and objects behave.
- Watching people work: Experienced workers wear a small head camera, with their consent, so Gallop can learn how the job is really done.
- Practice variations: Simulation generates variations of each example, so small changes on the line don't surprise the robot.
- Hands-on practice: An operator in a VR headset and gloves guides the robot through the task, on the same parts and with the same cameras as the real station.
- On-the-job learning: Once deployed, what the robot records feeds back into training.
The lesson so far is that the edge isn't more data but the right data: a real station, real parts, and real quality rules. Workers are the teachers, and their know-how is what Gallop is built on.
Inside the robot
Gallop works in four layers. A “planner” looks at the scene, understands the job, and breaks it into steps. A “supervisor”, a second AI, checks every step and re-plans when something goes wrong. A skills layer handles reaching, grasping, and placing. A control layer moves the arms and hands precisely and stops instantly if anything is wrong.
The layers talk through fixed interfaces. Because new AI models arrive every few months, this design lets Asus swap in a better one without rebuilding the robot.
Everything runs on a single NVIDIA Jetson AGX Thor computer inside the robot, using standard robotics software. Asus designs the robot body and the hand itself. The hand is the hardest part of any humanoid and is still scarce worldwide, so Asus builds its own in-house, with a sense of touch. Motors and other standard components come from established partners. Today, Gallop works from a fixed base with two arms, and a wheeled version comes next. Its strength is pick-and-place work at a station.
Key Features at a Glance
| Feature | Details |
|---|---|
| Target performance | 99.5% success within 20 seconds – the same standard Asus holds a person to. |
| Dexterous hand | Built in-house, with a sense of touch – handles small, thin, soft parts that a simple gripper can't. |
| Four-layer software | Planner, supervisor, skills, and control – errors get caught, the plan is revised, and the arms stop instantly if anything is wrong. |
| Swappable AI models | Fixed interfaces between layers – a better model can replace an old one without rebuilding the robot. |
| Onboard computing | A single NVIDIA Jetson AGX Thor inside the robot, running standard robotics software. |
| Existing workstations | Two arms and dexterous hands – works at the same stations and tools, so no factory rebuild. |
Availability and Roadmap
The Gallop program started in early 2026. Over the coming year, Asus plans to bring its own dexterous hand into the robot, put Gallop on wheels, and show it to customers at Computex 2027. Customer pilots are planned for 2027, and Asus is targeting the second half of 2028 for market availability.
Electronics manufacturing comes first, because it is the industry Asus knows best. Since many jobs share the same pick-and-place skills, what Gallop learns in one place can carry over to the next, which is how Asus sees it moving into hospitals and service sites over time.
Asus says Gallop is not meant to replace workers. It is designed for the repetitive, physically demanding stations that are hardest to staff, while the people who teach it move into setting up, supervising, and improving the robots.
Most Read
