The answer to human touch was never the eyes or ears. It was the Pacinian corpuscle, a layered sensor deep under the skin that acts like a natural filter, ignoring gentle strokes and static pressure while responding to impact, sudden pressure change and high-frequency vibration. These receptors cluster in palms and soles, the points where humans meet and strike the world.
More than a century later, Dr Xu Jincheng, a Waseda University graduate, named his 2021 company PaXini AI after that corpuscle, building high-precision tactile sensors and an embodied-perception stack that replicates roughly 100 million years of biological sensing logic for robots.
From research to scale
With the launch of its PX6AX GEN4 product line built on the GEN4 FUSE true 6D tactile chip, an upgraded headquarters and nationwide footprint, a joint-stock restructuring and a new RMB 1 billion funding round, PaXini says it has moved from technology accumulation into scaled industrial deployment.
Its ONE FOR ALL physical-AI showroom, nearly 3,000 square metres on the 39th floor of a central tower, is a working testbed rather than a prototype display. Walking inward past the 6-metre TORA-DOUBLE ONE giant humanoid, five zones lay out a full capability path: tactile chip, whole-body mechanics, data, dexterous operation and multi-robot coordination.
The GEN4 FUSE is described as the world’s first and only mass-produced native true 6D tactile sensing array chip. Around it, PaXini builds a whole-body mechanics suite covering fingertips, palms, skin, joints and soles. The flagship PXCap Pro data glove fuses a multi-array tactile sensor, a self-developed micro angle encoder and a wide-angle wrist camera for zero-drift pose data and high-fidelity ground truth.
The PIE framework
PaXini organises its stack as PIE: Perception, Intelligence, Execution. Perception gives the robot real feedback on contact, force, slip and deformation. Intelligence turns one physical interaction into a capability that can be learnt, reused and transferred. Execution takes intelligence off the screen and into real tasks through dexterous hands and robots.
The industry’s well-known gap is ‘good to watch, hard to deploy’. Humanoids that shine in labs falter in unstructured physical spaces, over-rely on vision, and keep hardware, data and execution as separate chains with no reliable physical feedback. PaXini answers with a full-stack loop.
To cut cost, it applies semiconductor wafer process to tactile-chip volume production, replacing custom precision parts so high-precision touch becomes base configuration rather than a premium option. To fix the data bottleneck, its OmniSharing DB auto-cleans, calibrates and labels multimodal data, building a collection-training-feedback flywheel.
The BYD test
The industrial proof is a recent BYD partnership. BYD runs the largest, densest and most demanding vehicle-assembly cluster in automotive history, and it wants a productivity layer that can read steel, paint, wiring and sealant. PaXini was embedded directly into BYD’s core stations as the perception and data core of final assembly, turning complex manual steps into deployable, efficiency-raising industrial productivity.
Humanoid competition has moved past appearance specs and demo videos. The dividing line is whether a robot can adapt to a complex physical world, complete real interactions and reach scaled commercial use. PaXini’s name carries an X for infinite interaction potential, and a bet that humanoids move from tech demo to scaled industry.
Editor’s note: This is an adapted translation of the original OFweek Robotics report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-09/ART-898890-8120-30702615.html.