In a nondescript industrial park in Tianjin’s Hexi district, a factory runs around the clock collecting data that will teach the world’s robots how to touch. The facility, operated by PaXini Technology (Tianjin) Co, is the world’s largest embodied intelligence data collection and model training base, and the company behind it has achieved something that seemed implausible just a few years ago: it has made tactile perception cheap enough to be a standard feature of industrial robotics rather than a laboratory luxury.
The story of how PaXini got there is documented in a report by China Daily, and it is a case study in the kind of patient, component-level innovation that is increasingly defining China’s approach to the embodied intelligence race. Tactile sensors, the devices that allow robots to sense pressure, texture, and posture through what the industry calls “tactile fingertips,” have been identified by China as one of the key critical chokepoint technologies, placed in the same category as lithography machines and advanced chips. The designation reflects a recognition that without the ability to sense the physical world with precision, robots cannot perform the dexterous manipulation tasks that would make them genuinely useful in factories, hospitals, and homes.
From 100,000 Yuan to 199 Yuan
“Five or six years ago, similar tactile sensor products cost as much as 100,000 yuan each, severely restricting the development of the entire robot industry,” recalled Nie Xiangru, co-founder of PaXini. The high import costs and the market’s domination by foreign suppliers made tactile perception inaccessible to all but the most well-funded research programs, creating a bottleneck that constrained the entire robotics ecosystem.
Breaking that bottleneck took the company four to five years of sustained R&D, culminating in what it describes as the world’s first 6D hall array multidimensional tactile sensor, a device that simultaneously measures force and torque across all six degrees of freedom. The commercial outcome has been striking: the entry-level sensor now retails at 199 yuan, a figure that represents just 0.2 percent of the price of comparable imported products. That price point has driven adoption at scale, today, eight in ten humanoid robots manufactured worldwide incorporate PaXini’s sensors, a market penetration that reflects both the technology’s performance and the straightforward economics of a product priced at a fraction of its foreign alternatives.
The cost reduction was not achieved through a single innovation but through a systematic approach to every layer of the technology stack. “The advancement of software algorithms enables us to reduce our reliance on hardware,” Nie explained. “Coupled with the growth of the industry, when prices become cheaper, the order volume increases, and with more orders, the costs can be further reduced. Naturally, the prices spiral downward.” The virtuous cycle of volume, cost reduction, and further volume is the same dynamic that drove the commoditization of solar panels and electric vehicle batteries, and PaXini is betting that tactile sensors will follow the same trajectory.
The Super EID Factory
The Super EID Factory, EID standing for embodied intelligence data, represents the next phase of PaXini’s strategy. The facility, which commenced operations last year in the Kongtian digital industrial park, as humanoid robots have begun deploying on consumer electronics assembly lines across China, runs around the clock collecting the tactile training data that current robotics datasets lack. Its daily throughput is equivalent to between 1.02 and 1.52 times a full worker’s career worth of data, and across a full year it will generate close to 200 million individual pieces of high-quality embodied intelligence training material. The factory encompasses thousands of tasks across full-chain embodied intelligence application environments, including automotive manufacturing, household tasks, office work, catering, supermarkets, and healthcare.
“For robots to be deployed effectively, they require sufficient scenario-based data support,” said Nie. The factory’s purpose is to generate the kind of rich, diverse, tactile-modal training data that current robotics datasets lack, data that captures not just what a robot sees but what it feels when it handles objects in varied real-world conditions. Lin Tong, director of PaXini’s data collection center, described the factory’s output as “a revolutionary solution to the core problems faced by the current physical intelligence industry, such as the scarcity of data sets and the lack of tactile modal data.”
Tianjin’s role as a free trade zone and the first city to release enterprise data classification standards and data cleaning operation lists has made it an attractive base for PaXini’s data operations. The city’s cross-border data transaction framework allows PaXini to collect and circulate data in ways that would be more restricted elsewhere in China, and the company plans to expand its data collection bases beyond Tianjin as it scales.
The Broader Significance
PaXini’s story illustrates a pattern that is becoming increasingly common in China’s technology sector: a domestic startup identifies a critical dependency on foreign technology, invests in the R&D to eliminate that dependency, achieves cost parity and then cost dominance through scale, and ends up with a globally competitive product that reshapes the market it entered. The tactile sensor market is a microcosm of the broader competition in embodied intelligence, and PaXini’s position at the center of it, with 80 percent of humanoid robots worldwide relying on its sensors, suggests that the outcome of that competition may already be more determined than it appears.