ASUS published details of Gallop on November 13, 2026, describing a humanoid robotics project whose first task is inside its manufacturing operation. The account concerns packing motherboard accessories. The crucial qualification is that 99.5% success within 20 seconds is a target, rather than an independently verified result or an established record.
Current configuration and future plans
According to ASUS, the current robot is a two-armed torso on a fixed base, using NVIDIA Jetson AGX Thor. Customer pilots are planned for 2027, with market availability targeted for the second half of 2028. These are the manufacturer's plans. They do not mean that the product is already available to purchase or operating at external customer sites.
Kaldata brought attention to the project. Its subheading presents the 99.5% figure in language that suggests achieved accuracy, while its body describes a target. Checking the original announcement resolves that discrepancy: the number describes the desired workstation standard. SIExpo is consequently not presenting it as a record.
Compute capability is not workstation performance
NVIDIA positions Jetson Thor as a platform for physical AI and robotics. Such computing can support models and sensor processing, but its presence does not measure grasping success, packing quality or safe recovery from an error. A processor specification and a production-station result are different kinds of evidence. Neither can stand in for the other.
Our technical interpretation is that the task needs to be evaluated as a sequence. The robot must identify an appropriate object, grasp it, transfer it and place it correctly. With small packaged accessories, failure at one stage can invalidate correct execution elsewhere. A visually impressive individual movement does not establish that a finished box is correct.
The same distinction applies to time. If the fastest demonstration fits the desired cycle but an operator needs a long intervention after a failure, that demonstration does not describe average throughput. An evaluation would need cycle duration, success, human intervention and downtime together. These are suggested assessment criteria, not unpublished measurements from Gallop.
What a useful test would establish
We would first distinguish the identity and number of accessories from the appearance of the completed package. The next question is whether omissions are detected, rather than whether selected successful attempts look convincing. Predefined scenarios with varying arrangement, packaging reflections and partial occlusion could be informative. They should represent the actual task rather than being selected after a successful run.
Recovery deserves its own measurement. Does the system recognize that an operation has not finished? Can it stop safely and hand the situation to a person with an intelligible explanation? Reporting those outcomes would make comparisons between approaches more useful, without treating a humanoid shape as evidence of operational readiness.
A further assessment would preserve the test conditions. Changing objects, lighting or acceptance criteria between runs makes a headline percentage difficult to interpret. Reporting what remained fixed, what varied and how failed cases were counted would let another team understand what was demonstrated. This is an editorial recommendation for evaluating the project; it is not a claim that ASUS has or has not followed a particular unpublished protocol.
The development is therefore the concrete factory project and its stated quality objective. The next useful evidence for external observers would be a sustained-operation report with explicit test boundaries. ASUS's described configuration and roadmap do not establish readiness for unrelated manufacturing tasks. SIExpo has neither observed the factory tests in person nor performed its own comparative evaluation.
The cover is a conceptual AI illustration of robotic packing, not an exact depiction of Gallop.
Editorial verification: 2026-11-14T17:12:57.436197+00:00.
