Shenzhen’s robotics supply chain has spent two years selling movement. The next layer sells touch, and it is being built by people who were in school when the first humanoid prototypes walked.
YuanSheng, a Shenzhen company founded in 2025 by Oxford-trained physicist Hou Ningzhe, born in 1999, is assembling a hardware stack around one problem: a robot that can see a part but cannot feel it will drop it. Its product line runs from multi-dimensional tactile sensors to data-collection gloves to electronic skin, and now into automotive-grade manufacturing.
Why vision runs out
Industrial assembly is the test case. The robot has to locate the part, approach it, make contact and complete the insertion. Cameras give shape and coordinates. They do not give the millimetre-level stall in a connector, or the slip of an object rotating inside a grip, because the fingers occlude exactly the region where that information lives.
Humans close that loop with vision, touch and muscle control together. Machines need force and contact data as a separate channel, which makes tactile sensing a functional requirement rather than a specification line item. Rigid, standardised pick-and-place can survive without it. Flexible objects, precision insertion and unstructured environments cannot.
The hardware underneath
The TACTA sensor, built on semiconductor processes, is under 4 millimetres thick and samples at 1,000 hertz, fast enough to catch slip and impact events that last milliseconds. The electronic skin version is 1.75 millimetres thin and, on published figures, covers close to 80 per cent of a humanoid’s body surface with roughly 18,000 contact points.
The Magic Glove Concept, launched globally at the World Robot Conference in August 2026, is the data play. It uses split modular construction to fit different hand sizes and skeletal structures, recording motion trajectories and contact mechanics at the same time, then mapping that human data onto a robot hand whose joints do not match the operator’s.
That mapping problem is the bottleneck most data-glove vendors understate. A glove that records beautifully and retargets badly produces a dataset that trains the robot to fail in a slightly different way.
Borrowing the car industry’s discipline
In August the company formed a joint venture with Shangfu Dianji to build an automotive-grade piezoresistive tactile sensor line, adopting IATF 16949 quality management and automated production control. That is a deliberate move: batch consistency, not peak performance, is what stops a sensor from becoming a reliability problem at 10,000 units a month.
Doing all of this inside one city is the Shenzhen argument. Semiconductor processing, precision structures, algorithms, manufacturing and now automotive electronics sit within commuting distance, which is why the shift from consumer electronics to embodied intelligence is happening here faster than in cities with stronger research brands.
Touch is the least photogenic layer of the humanoid stack. It is also the one that decides whether the machine on the poster can hold a job.
Editor’s note: This is an adapted translation of the original Shenzhen News report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.sznews.com/news/content/2026-11/02/content_32206155.htm.