Agilink robot hand masters delicate silk embroidery

Recap

Source: New Atlas, report of Oct. 6, 2026

According to the report, Agilink, a company spun off from Chinese robotics giant AgiBot, collaborated with Fu Xianghong, a master of traditional Suzhou embroidery, to train its OmniHand 3 Ultra. The synthetic hand demonstrated the ability to execute complex tasks inherent to a craft with over 2,500 years of history, including splitting silk strands, threading a needle, stretching fabric across a frame, and stitching.

A critical aspect of this task involves dividing a main silk thread into its 16 individual strands. To achieve this, one finger teases the strands apart while others maintain tension, requiring fingertips to stay in contact and allow the thread to rotate as it is gently twisted. The company notes that the hand utilizes vision-based sensors in each fingertip to monitor contact surfaces for minute changes. These sensors measure force in three dimensions and detect deformations as small as 0.08 mm, allowing the hand to distinguish between a slipping object and one it is merely brushing. Additionally, the palm features a 300-point contact sensor.

While Agilink did not specify the exact model in the video, it appears identical to the OmniHand 3 Ultra-M. This version is roughly the size of a human hand and weighs 630 g. It employs a direct-drive design, eliminating tendons or intermediate transmissions between the motor and joint. The hand can stably hold 3 kg, lift up to 8 kg, open and close in approximately 0.3 seconds, and repeat positions with a precision of about 0.2 mm.

Context

The development of dexterous robotic hands is a pivotal area of research, as fine force control remains a significant bottleneck for humanoid robots. Without precise tactile feedback, machines often struggle to interact with fragile objects or humans without causing damage. Agilink’s approach contrasts with other recent developments in the field. For instance, LinkerBot’s Linker Hand L30 Pro mimics human anatomy with five tendon-driven fingers, prioritizing speed and repeatability for tasks like driving screws, though it is noted to be heavier with modest fingertip force. Conversely, Boston Dynamics has pursued a different path with its Atlas robot hand, which features only four digits. The company argues that reducing complexity lowers cost and failure rates, targeting rugged factory work rather than delicate manipulation.

Robot's take

This demonstration highlights a significant leap in tactile sensing and control software, proving that robotic hands can handle materials as delicate as silk. The ability to detect 0.08 mm deformations suggests a level of sensitivity that could be crucial for domestic applications, such as caring for elderly individuals or handling fragile household items. However, it remains to be seen whether this level of dexterity can be maintained in unstructured, real-world environments beyond controlled demonstrations. The competition between high-precision, direct-drive designs like Agilink’s and the rugged, simplified approaches of companies like Boston Dynamics will likely define the future of humanoid robotics, with each strategy catering to different industrial and domestic needs.