PaXini Technology
Last reviewed
May 16, 2026
Sources
20 citations
Review status
Source-backed
Revision
v5 · 4,181 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 16, 2026
Sources
20 citations
Review status
Source-backed
Revision
v5 · 4,181 words
Add missing citations, update stale details, or suggest a clearer explanation.
| PaXini Technology | |
|---|---|
| General information | |
| Full name | PaXini Perception Technology (Shenzhen) Co., Ltd. |
| Chinese name | 帕西尼感知科技 |
| Founded | April 2021 |
| Founder and CEO | Xu Jincheng (Hsu Jincheng) |
| Headquarters | Shenzhen, Guangdong, China |
| Additional offices | Shanghai, Tianjin |
| Industry | Robotics, tactile sensing, embodied AI |
| Products | Tactile sensors, dexterous hands, humanoid robots, data platforms |
| Valuation | Over $1.45 billion (March 2026) |
| Total funding | Over CNY 2 billion (approximately $280 million) |
| Key investors | JD.com, BYD, SAIC Motor, Meta, Huangpu River Capital, TCL |
| Website | paxini.com |
PaXini Technology (Chinese: 帕西尼感知科技, PaXini Perception Technology) is a Chinese robotics company headquartered in Shenzhen, Guangdong that specializes in high-precision multidimensional tactile sensors, dexterous robot hands, and humanoid robots. Founded in April 2021 by Xu Jincheng (also romanized Hsu Jincheng), a former doctoral student of robotics pioneer Shigeki Sugano at Waseda University in Tokyo, PaXini is widely cited as China's largest tactile sensor maker and the only Chinese company commercially manufacturing a Multi-Tactile Dexterous Hand at scale.[1][2]
The company's name derives from the Pacinian corpuscle, one of the four mechanoreceptors located beneath human skin that detects deep pressure and high-frequency vibration. This naming reflects PaXini's central thesis that the missing modality for embodied AI is not vision or language but a digitized sense of touch. As of early 2026 PaXini has raised a cumulative total of over CNY 2 billion (approximately $280 million) across angel, pre-A, A, A+, and Series B rounds, achieving a post-money valuation exceeding CNY 10 billion ($1.45 billion) after the March 2026 Series B led by Huangpu River Capital and Kaitai Capital.[3][4][5]
PaXini operates a vertically integrated stack that it describes as a "physical perception infrastructure" for embodied intelligence, spanning four layers: tactile sensors, a robot platform, an embodied data factory, and end-customer scenario applications. Its flagship products include the PX-6AX-GEN3 multidimensional tactile sensor, the PX6D and PXTS Hall-effect six-dimensional force and torque sensors, the DexH13 four-finger dexterous hand, the TORA-ONE and TORA-DOUBLE ONE humanoid robots, and the OmniSharing DB dataset generated by the world's first omni-modal embodied AI data factory in Tianjin.[6][7]
PaXini was founded in April 2021 in Shenzhen by Xu Jincheng. Xu was born in 1989 in Taichung, Taiwan, and completed his mandatory military service before traveling to Japan for graduate study. At Waseda University in Tokyo he earned his doctorate in the Sugano Laboratory, the research group led by Professor Shigeki Sugano. The lab is historically significant because Waseda University is widely credited with developing WABOT-1, the world's first full-scale anthropomorphic robot, in the 1970s, giving the Sugano Lab a continuous lineage in humanoid robotics research that spans half a century.[1][8]
During his doctoral work Xu specialized in tactile sensing and human-robot physical interaction, areas of long-standing strength for Sugano's group. By 2020 Xu had become convinced that progress in deep learning and the public success of systems such as AlphaGo had created the algorithmic preconditions for embodied AI, but that the field still lacked an industrialized tactile-sensing layer comparable to the cameras that vision models could rely on. Lu Qi, the founder of MiraclePlus (the rebranded Y Combinator China program) and a former president of Baidu, met Xu while still in Japan and offered angel funding.[1][2]
In April 2021 Xu flew from Tokyo via Shanghai to Shenzhen and registered the company. He chose Shenzhen partly because supplier turnaround in the Greater Bay Area allowed prototype iteration in one to two weeks rather than the three months he had encountered in Japan, and partly because Shenzhen's electronics manufacturing ecosystem could absorb large volumes of low-cost sensors once the company moved into mass production. The choice of the name Pacini, anglicized to PaXini, was deliberately academic: Xu wanted the corporate brand to evoke the biological mechanoreceptor whose function the company aimed to digitize.[2][8]
During its first three years PaXini concentrated on driving down the unit cost of a high-precision tactile sensor while simultaneously improving its measurement bandwidth. According to disclosures around the Series B round, PaXini reduced the per-unit cost of a multidimensional tactile sensor from roughly CNY 100,000 to as little as CNY 199, a roughly 500-fold cost reduction that was decisive for placing thousands of sensors on a single humanoid robot at viable bill-of-materials economics.[3]
The core product line during this period was the PX-6AX series of multidimensional tactile sensors. Each sensor offered 0.01 N force resolution across its full measurement range and produced up to 15 simultaneous sensing dimensions, including six-axis force, surface texture, and material elasticity. Adoption of these sensors spread from research groups into automotive assembly, consumer electronics manufacturing, and warehouse logistics. By 2024 sensor shipments had grown substantially, with the bulk of volume destined for factory automation cells and a growing number of humanoid robot integrators. PaXini also reported entry into the supply chain of Apple for tactile components used in manufacturing applications.[3][9]
Between 2021 and 2024 PaXini completed angel, pre-A, and Series A funding rounds that drew an unusually wide group of corporate investors. Lu Qi's MiraclePlus and the venture arms of automotive companies including BYD, SAIC Motor, and BAIC Group participated in early rounds. According to subsequent disclosures, BYD became one of PaXini's largest single shareholders with a 13.4 percent stake. TCL Capital and Addor Capital also joined as financial investors. The participation of three Chinese automakers in the equity of a tactile-sensor company underlined how seriously the domestic auto industry took the prospect that humanoid robots and dexterous hands would soon enter electric vehicle assembly lines.[10][11]
On 5 August 2025 PaXini disclosed a Series A round of CNY 1 billion (approximately $139 million) closed in four months and led by JD.com, the Chinese e-commerce group. New participants included Puxin Capital and XGD, a payment-terminal manufacturer, while existing investors TCL Capital and Addor Capital extended their commitments. The Series A was JD.com's sixth robotics investment within the preceding three months, following positions in AgiBot, Spirit AI, EngineAI Robotics, LimX Dynamics, and RoboScience, and reflected an intensifying contest among the largest Chinese internet companies, including Alibaba and Meituan, for stakes in the embodied AI supply chain.[10][11]
JD.com Senior Vice President He Xiaodong described the investment as a long-term strategic commitment, framing PaXini's tactile stack as foundational infrastructure for warehouse, logistics, and human-robot collaboration applications that JD itself was preparing to deploy. The proceeds were earmarked for expansion of the Super EID Factory in Tianjin and acceleration of the DexH13 program. The Series A also funded PaXini's first sustained overseas marketing campaign, including its participation in iREX 2025 in Tokyo and its CES 2026 debut in Las Vegas.[10][11]
On 23 June 2025 PaXini opened the PaXini Super EID Factory (Super Embodied Intelligence Data Factory) in Tianjin Municipality. The facility was described by Chinese state media as the world's largest embodied AI data factory, comprising approximately 12,000 square metres of floor space, 150 standardized data acquisition cells, and a planned annual production of nearly 200 million high-quality multimodal training samples. The opening was attended by representatives from PaXini's automotive and internet investors and was framed as the physical anchor of the OmniSharing DB dataset.[12][13]
The factory uses a 15+N scenario matrix that pre-defines fifteen canonical task families across automotive manufacturing, 3C electronics assembly, household, office, medical care, retail, and food-service environments, with an open-ended N category for customer-specific scenarios. Inside each cell, human operators wear PaXini's proprietary PMEC (PaXini Motion and Encounter Capture) suits, gloves, and headsets that capture body motion, hand kinematics, and high-resolution tactile signals at the contact surface. The captured trajectories are then retargeted onto multiple robot embodiments through a software pipeline so that the dataset can be reused as humanoid hardware evolves.[12][13]
On 6 March 2026 PaXini announced a Series B round of more than CNY 1 billion ($145 million) that took its post-money valuation above CNY 10 billion (approximately $1.45 billion), confirming its status as a unicorn. The round was co-led by Huangpu River Capital, Kaitai Capital International Group, and the state-backed CIM International Group Inc.[3][4]
The full Series B roster included Huangpu River Capital, Kaitai Capital International Group, Xin'an Capital, Guangzhou Emerging Fund, Zhilai Capital, LEO LION, Nanling Fund, Xiangcheng Financial Holding, Addor Capital, Uni-Trend, Meta, BYD, JD.com, TCL, and Zhuhai Sci-Tech Industry Group, the latter associated with Huafa and Gree Electric. Meta's participation was widely noted as evidence that a leading U.S. technology company viewed PaXini's tactile stack as strategically important enough to take an equity position despite the broader U.S.-China technology tensions.[4][5]
PaXini operates a four-layer product hierarchy that it markets as a closed loop: tactile sensors, robot platforms, embodied data, and scenario applications. Each layer is sold both as a standalone product and as part of an integrated system.
The PX-6AX-GEN3 is PaXini's third-generation multidimensional tactile sensor and the foundation of its product line. It is built around a six-dimensional Hall-effect sensing array that detects deformation in a structured elastomer pad in three translational and three rotational axes, allowing the sensor to recover not only normal pressure but lateral shear, torque, and contact geometry.
| PX-6AX-GEN3 specifications | Value |
|---|---|
| Sensing dimensions | 15 simultaneous (6-axis force, texture, elasticity, slip, contact area) |
| Force resolution | 0.01 N across full measurement range |
| Repeatability | Better than 0.5 percent of full scale |
| Output rate | 1,000 Hz |
| Product family | 3 product series, 12 models |
| Reference unit price | From $49 |
The PX-6AX-GEN3 was introduced at iREX 2025 in Tokyo in December 2025 and updated for CES 2026. It is sold as fingertip pads, palm pads, full-finger sleeves, and large-area skin tiles.[6][14]
PaXini introduced the PX6D and PXTS at CES 2026 as the world's first commercial Hall-effect six-dimensional force and torque sensors designed specifically for embodied AI. Unlike traditional steel strain-gauge load cells, the PX6D uses an advanced polymer elastomer paired with a Hall-effect array. PaXini cites three advantages: lower unit cost, much lighter weight that makes whole-body coverage practical on a humanoid robot, and reduced sensitivity to aging and creep. The PX6D and PXTS form the core of PaXini's full-body force-sensing initiative, which aims to distribute six-axis force measurement across joints, limbs, and torso rather than concentrating it at the wrist.[6][14]
The DexH1 was PaXini's first commercial dexterous hand and established the company's reputation in bionic manipulation. The hand was designed to mimic fine human motor movements and shipped to research laboratories and humanoid-robot integrators during 2024 and 2025. Although superseded by the DexH13, the DexH1 continues to be offered for budget-sensitive deployments.[10]
The DexH13 is PaXini's flagship dexterous hand and its highest-profile product after the tactile sensor line. It was launched globally at CES 2026.
| DexH13 specifications | Value |
|---|---|
| Configuration | 4 fingers, bionic layout |
| Degrees of freedom | 16 |
| Mass | Approximately 700 to 1,000 grams |
| Payload | 5 kg |
| Tactile units | 1,140 ITPU multidimensional tactile processing units |
| Vision | Integrated AI camera |
| Actuation | Hollow-cup motors |
| Communication | EtherCAT, Modbus |
| Reference price | $17,200 |
The DexH13 covers its fingertips, finger pads, and palm with PaXini's ITPU (Intelligent Tactile Processing Unit) modules, allowing it to sense object shape, pressure distribution, material compliance, and slip in real time. PaXini's CES 2026 demonstrations showed the DexH13 grasping irregular objects including test tubes, cubes, and spheres, mirroring human gestures, and performing delicate fine-motor tasks such as turning knobs. The hand's communication stack supports the EtherCAT and Modbus protocols, allowing it to drop into existing industrial robot cells.[6][15]
The TORA-ONE is PaXini's tactile humanoid robot and the showcase platform for full-stack integration of the company's sensing technology.
| TORA-ONE specifications | Value |
|---|---|
| Height | 146 to 186 cm (adjustable) |
| Total degrees of freedom | Up to 53 |
| Body degrees of freedom | 21 |
| Hand configuration | Four-fingered bionic hands |
| Hand degrees of freedom | 26 |
| Tactile sensors | 1,956 sensors across 7,824 channels |
| Sensing dimensions | Multidimensional (six-axis force, texture, elasticity) |
TORA-ONE gained considerable public attention at CES 2026 in Las Vegas, where it autonomously executed a full ice-cream-making workflow at the Enterprise AI Zone in North Hall. The demonstration required the robot to manipulate dispensing levers, measure ingredients, pick up wafer cones, dispense ice cream, and hand finished cups to human visitors, each step requiring distinct grip force and fine alignment.[6][15][16]
The TORA-DOUBLE ONE (also written DoubleOne) is PaXini's universal general-purpose humanoid robot, designed for industrial and service deployments.
| TORA-DOUBLE ONE specifications | Value |
|---|---|
| Height | 168 cm (foldable design) |
| Mass | 70 kg |
| Payload | 6.5 kg |
| Maximum speed | 6.5 km/h |
| Total degrees of freedom | 58 |
| Hand configuration | Five-finger hands, 18 DOF per hand |
| Runtime | Up to 6 hours per charge (swappable battery) |
| Computing | NVIDIA Jetson AGX Orin and x86 real-time controller |
| Perception | RGB, RGB-D, 3D LiDAR, depth sensors |
| Connectivity | Wi-Fi, Bluetooth, 5G |
| Reference price | $45,000 |
TORA-DOUBLE ONE adds a wheeled or articulated base, swappable batteries, and a more powerful compute stack to the TORA-ONE design. Live demonstrations at iREX 2025 and CES 2026 showed it crossing stepped obstacles and manipulating pallet jacks in logistics scenarios. Its modular hand interface lets operators swap between DexH13 hands and simpler grippers depending on the task.[17][18]
OmniSharing DB is PaXini's omni-modal embodied AI dataset, produced inside the Super EID Factory in Tianjin and distributed through a planned global cloud service. Each data record bundles synchronized streams of body motion capture, hand kinematics, RGB and depth video, and six-axis tactile traces, all referenced to the contact geometry captured by the PMEC suits. PaXini designed the dataset to address what Xu calls industry-wide "data anxiety": the chronic shortage of high-quality contact-rich data that is the principal bottleneck for vision-language-action models in dexterous manipulation. The factory targets nearly 200 million records per year and PaXini has stated that the dataset will be sold or licensed to overseas customers as a separate product line.[3][12][13]
The centrepiece of PaXini's technology is a six-dimensional Hall-effect tactile sensing architecture. A structured elastomer skin is embedded with a regular array of small magnets above a substrate populated with Hall sensors. When the skin is loaded, the magnets shift in three translational and three rotational dimensions, and the changes in magnetic flux are decoded into six-axis force, slip, vibration, and contact-area signals. The approach allows PaXini to manufacture sensors at low cost using standard PCB and elastomer processes while preserving high spatial resolution and stable response.[6][14]
The full-stack design choice that distinguishes PaXini from competitors using purely optical or capacitive sensing is its commitment to whole-body force sensing. By scaling the PX6D and PXTS sensors down to low cost and low mass, the company can populate the entire surface of a humanoid robot with six-axis sensing rather than concentrating force measurement at the wrist. The TORA-ONE's 1,956 sensors covering 7,824 channels is an early implementation of this distributed sensing approach, and PaXini has signalled that future TORA generations will extend the coverage further into the torso and lower body.[7][14]
PaXini bundles its sensor arrays with an on-board processing layer it calls the Intelligent Tactile Processing Unit (ITPU). Each ITPU module performs local signal conditioning, calibration, slip detection, and feature extraction so that the host robot controller receives semantically meaningful tactile events rather than raw analogue traces. The DexH13's 1,140 ITPU modules per hand allow the robot to react to slip and unexpected contact with sub-millisecond latency without saturating its main compute bandwidth.[6][15]
PaXini's data philosophy differs from the teleoperation-heavy approach used by many North American humanoid programs. Rather than collecting data by piloting robots, PaXini collects motion and contact data directly from humans wearing the PMEC suit and then retargets the trajectories to multiple robot embodiments through a software layer. Xu has argued publicly that this approach yields data that survives hardware redesigns and that humans, unlike robots, do not require expensive supervision during data collection. The retargeting pipeline is one of the principal pieces of intellectual property protecting the OmniSharing DB asset.[7][12]
At CES 2026 PaXini articulated its strategy as a closed loop of four layers: Tactile Sensors -> Robot Platform -> Embodied Data -> Scenario Application. The argument is that each layer reinforces the others. Sensors generate data, data trains models, models drive better robot platforms, and robot platforms deliver scenario-specific applications that in turn motivate new sensor designs. Xu describes this stack as "physical perception infrastructure" and has compared its role in embodied AI to the role of the GPU stack in generative AI.[6][7]
PaXini divides its commercial markets into five primary verticals, each addressed by a combination of sensor-only sales, dexterous-hand sales, and full humanoid deployments.
| Vertical | Representative use cases |
|---|---|
| Automotive manufacturing | Final assembly, fastening, inspection, quality assurance |
| 3C electronics and consumer | Component handling, screw driving, board assembly |
| Logistics and warehousing | Picking, packing, palletizing, pallet-jack operation |
| Medical and healthcare | Specimen handling, lab automation, assistive devices |
| Retail and food service | Order fulfillment, ice cream and beverage preparation |
Automotive customers include investors BYD, SAIC Motor, and BAIC Group, who use PaXini sensors and hands in pilot lines for electric vehicle assembly. Logistics applications are driven by JD.com, which operates one of the largest warehouse networks in China. Consumer-facing demonstrations such as the TORA-ONE ice cream booth at CES 2026 are intended in part to qualify the platform for retail and food-service contracts in the United States, Japan, and South Korea, which Xu has named as priority overseas markets.[3][10][16]
Xu has stated that he expects industrial humanoid deployments at meaningful scale to materialize within two to three years, contingent on continued progress in the underlying vision-language-action models that PaXini's data platform is designed to train.[7]
PaXini began direct overseas marketing in late 2025. Its first independent international showcase was iREX 2025 in Tokyo in December 2025, where it demonstrated TORA-ONE preparing ice cream for attendees and TORA-DOUBLE ONE navigating stepped obstacles. The company also opened flagship stores on AliExpress and Amazon during the iREX period, signalling a direct-to-developer distribution channel for its sensor and hand product lines.[14]
Its CES 2026 debut in January 2026 in Las Vegas was the company's first significant U.S. exhibition. PaXini occupied space in the Enterprise AI Zone of the North Hall and ran live demonstrations of the DexH13, TORA-ONE, and PMEC data acquisition. The company has publicly identified the United States, Europe, Japan, and South Korea as priority markets and has built a separate global sales channel (sales_global@paxini.com) at its Shanghai office.[6][7]
PaXini operates at the intersection of two competitive markets: tactile sensors and dexterous robotic hands. Its competitive position is unusual because it is one of the few firms that operates across both layers and into full humanoid systems.
| Company | Country | Primary overlap with PaXini |
|---|---|---|
| Shadow Robot Company | United Kingdom | Five-finger dexterous hands for research |
| Wonik Robotics (Allegro Hand) | South Korea | Four-finger dexterous hands, academic market |
| Schunk | Germany | Industrial grippers and force sensing |
| Inspire Robotics (因时机器人) | China | Multi-finger dexterous hands for humanoids |
| Sanctuary AI | Canada | General-purpose humanoid with dexterous hands |
| Tesla Optimus hand team | United States | In-house dexterous hand for Tesla humanoid |
| GelSight | United States | Optical tactile sensors |
| ATI Industrial Automation | United States | Strain-gauge six-axis force/torque sensors |
Against Western incumbents such as Shadow Robot, Schunk, ATI, and GelSight, PaXini's principal differentiator is cost. The reduction of multidimensional tactile sensor pricing from CNY 100,000 to CNY 199 referenced at the Series B announcement is several orders of magnitude below the prices charged by Western competitors for comparable measurement specifications, and is what makes whole-body sensor coverage on a humanoid robot economically feasible at all. Against Asian competitors such as Inspire Robotics in China and Wonik Robotics in South Korea, PaXini's differentiator is integration: it bundles the sensor, the hand, the robot, and the dataset into a single supply relationship, which is attractive to integrators that lack their own tactile R&D.[3][14]
In the broader humanoid robot field PaXini does not compete directly with locomotion-first firms such as Unitree, Booster, or Boston Dynamics, but is instead a candidate supplier to them. Its TORA platforms are nonetheless full-stack humanoids and therefore compete with general-purpose humanoid programs at Sanctuary AI, Figure AI, Apptronik, Agility Robotics, and 1X Technologies in the U.S., and with UBTech, AgiBot, Fourier Intelligence, and EngineAI in China.
| Round | Date | Amount | Lead and notable investors |
|---|---|---|---|
| Angel | 2021 | Undisclosed | MiraclePlus (Lu Qi) |
| Pre-A and A series corporates | 2022 to 2024 | Undisclosed | BYD, SAIC Motor, BAIC Group, TCL Capital, Addor Capital |
| Series A | August 2025 | CNY 1 billion (approx. $139 million) | JD.com (lead), Puxin Capital, XGD, TCL Capital, Addor Capital |
| Series B | March 2026 | Over CNY 1 billion (approx. $145 million) | Huangpu River Capital, Kaitai Capital, CIM International Group (co-leads); Meta, BYD, JD.com, TCL, Zhuhai Sci-Tech Industry Group, 9 others |
Total disclosed funding through Series B exceeds CNY 2 billion (approximately $280 million), and the Series B post-money valuation was placed above CNY 10 billion ($1.45 billion).[3][4][5][10]
Xu Jincheng (徐津诚), also romanized as Hsu Jincheng, serves as founder and chief executive officer of PaXini. Born in Taichung, Taiwan, in 1989, he completed mandatory military service in Taiwan before pursuing a doctorate at Waseda University in Tokyo in the Sugano Laboratory. His doctoral research focused on tactile sensing and human-robot interaction. Xu is also a member of the IEEE Robotics and Automation Society and has presented PaXini's work at international robotics venues including iREX, CES, and academic workshops on dexterous manipulation.[1][2][8]