AGILINK Releases OmniGlove Pro Data-Collection Glove; Dexterous Hand Completes 72-Process Production Line Validation

Author | Huang Nan

Editor | Yuan Silai

On October 29, 2026, at the AGIBOT Innovation Day · Dexterous Manipulation Sub-forum, AGILINK showcased key progress in bringing dexterous hands into real industrial working conditions.

At the POC validation workstation on the Shaoxing Wolong production line, the OmniHand dexterous hand completed factory whole-machine performance testing on dexterous hands of the same series, running through 72 consecutive processes—from material loading and screw fastening to self-inspection and unloading and return to position—with no need for real-time human intervention throughout. A single round of the entire process took 8.5 minutes, 30 seconds slower than the 8 minutes of a skilled worker, with a task success rate of 98%, and it can operate continuously for 8 hours.

The OmniHand dexterous hand performs factory whole-machine performance testing on dexterous hands of the same series (Image source/Company)

Among these, screw fastening posed the greatest challenge. The anti-loosening screws themselves wobble at the millimeter level, and combined with the self-weight of the screw gun and the recoil caused by cable pulling, a mere 1-millimeter deviation of the hand would amplify the deviation at the tip of the gun to 3 to 4 millimeters. During testing, the team also artificially introduced disturbances such as robotic arm pushing, fixture base displacement, and lighting changes; the dexterous hand relied on coordinated vision and touch to correct deviations in real time, and could automatically retract and re-align when a screw-in attempt failed.

The stability of the entire operation comes from real-machine reinforcement learning optimization of single-step manipulation capabilities, as well as the support of an industrial deployment agent framework. The latter is responsible for process verification, failure rollback, and state synchronization between workstations.

Shaoxing Wolong production line (Image source/Company)

AGILINK CEO Qiao Tianjie defines this production line test as a landmark node marking the dexterous hand industry's transition from the development state to the deployment state. "The development state is about how impressive the demo is; the deployment state assesses cycle time, success rate, stability, and ROI."

In his view, after decades of iteration, grippers can cover 70% to 80% of deterministic processes in industrial scenarios; the goal of dexterous hands is to tackle the remaining 20% to 30% of complex and delicate operations that grippers struggle to handle. Using a general-purpose dexterous end effector can adapt to multiple types of processes, reducing the retrofit costs and scenario limitations brought by dedicated machines for dedicated tasks on traditional production lines.

The production line POC validated the feasibility of the hardware and operation framework at real workstations, while also bringing the industry's long-standing data shortcomings to the fore.

AGILINK Chief Embodied Scientist Huang Bidan pointed out that most current dexterous manipulation model training relies on pure vision datasets, lacking real physical interaction information such as force feedback and tactile sliding. Models perform excellently in simulation but are highly prone to failure when transferred to real physical environments; tactile dimensions such as temperature, shear force, and contact deformation are difficult to generate through simulation and must be collected through real physical interaction.

The two mainstream data collection paths each have shortcomings: data collected through real-machine teleoperation is high in quality but expensive and extremely dependent on professional operators, making large-scale expansion difficult; first-person ego human-hand perspective data has a low acquisition threshold and broad scenario coverage, but relying solely on visual tracking lacks real tactile information, has limited precision, and is more suitable only for pre-training, insufficient to support high-precision delicate manipulation tasks.

In response to these pain points, AGILINK officially released the OmniGlove Pro data collection glove at the event and began mass production delivery, attempting to establish a new quality evaluation standard from the data source.

OmniGlove Pro data collection glove (Image source/Company)

Traditional data collection mostly uses the similarity of human hand motion replication at the visual level as the yardstick; the core logic of "real-machine Ready" is that a human hand completes an operation in a specific scenario, and after collection by the glove and back-end processing, the robot body can reproduce the complete task under the same scenario conditions—this serves as the core criterion for determining whether data is qualified.

At the hardware level, the fingertips of the OmniGlove Pro adopt a fusion scheme of EMF magnetic field positioning and IMU anti-interference, with high-density tactile sensing points arranged across the entire hand; pose and tactile signals maintain high-frame-rate output, and the weight of the entire glove is kept at a relatively light level. In the fabric glove form, EMF provides the fingertip positioning reference, and the full-hand tactile points are used to compensate for end-effector positioning errors caused by flexible deformation.

In terms of form factor selection, the team chose a fabric glove solution, which is less intrusive to daily operations and more suitable for long-duration, large-scale continuous collection. At the same time, AGILINK's self-developed OmniHand dexterous hand body is also equipped with a fabric tactile glove at the factory; the data collection glove and the execution body are aligned as much as possible in tactile point distribution and appearance, to reduce information loss when mapping data collected from humans to the robot execution end.

Official data shows that as of October 2026, cumulative mass production of AGILINK's OmniHand series dexterous hands has exceeded 15,000 units. Qiao Tianjie believes that mass production at the ten-thousand-unit level is the basic threshold for dexterous hands to move toward commercial deployment; only upon reaching this scale do supply chain delivery stability, whole-machine yield convergence, and cost reduction have realistic conditions.

AGIBOT Innovation Day · Dexterous Manipulation Sub-forum (Image source/Company)

Relying on continuous iteration across two real deployment lines—industrial and commercial services—AGILINK's dexterous hands have achieved regular operation in the commercial interaction scenario at Chimelong Zhuhai; the dexterous hand workstation at Wolong Electric Drive is expected to complete formal production line integration within the year.

From workstation validation on the Wolong production line to OmniGlove Pro establishing the "real-machine Ready" data collection standard, AGILINK is attempting to run through a closed-loop path: real deployment scenarios generate interaction data, high-quality collected data feeds back into model iteration, and the iterated models then unlock more new processes.

At the event, AGILINK announced that it would open tactile data, manipulation models, and the industrial deployment framework to body manufacturers, scenario integrators, and model R&D teams, expanding data supply and scenario coverage through ecosystem collaboration.