Author: Peng Zhao (Founder of Zhicifang, Co-Founding Partner of Yunhe Capital)
Just as people always synchronize their clocks to the same tick before executing a task requiring precise coordination, a robot's sensory systems must also "synchronize" before it can walk or grasp objects.
This year, Beiwai Sensing in Wuxi launched an embodied intelligence multi-sensor fusion pose navigation unit, integrating a 360° 3D LiDAR, an inertial/pose unit, a global shutter industrial camera, and an edge computing platform into a single chassis.
Before shipment, engineers first "synchronized" these three types of sensors. Through time alignment and extrinsic parameter calibration, they adjusted each sensor's internal clock to the same millisecond and calibrated their fields of view to the same spatial coordinate system, ensuring that point clouds, images, and inertial data recorded the same moment and orientation.
This process rarely appears on the big screens at product launches, yet it determines whether a machine remains merely an exhibit or can truly work on the factory floor.
Starting October 30, the 2026 World IoT Expo opened in Wuxi. Over the past few days, moving between exhibition halls and conference venues, humanoid robots and AI models remained the center of attention. Sensors are typically hidden inside robot joints or tucked away in corners of production lines, making them less visible to the naked eye.
However, during this IoT Expo, my primary focus was on engineering details like "clock synchronization" that lie behind the scenes.
After all, when AI enters factories, power grids, and roads, the first hurdle is whether the perceptual data it receives is reliable.
Changing the Exam: From Competing on Connection Scale to Competing on Perceptual Value
Last year, IoT Think Tank predicted that IoT would support AI's entry into the physical world.
A year later, this prediction has been validated by the industry.
At this year's IoT Expo, embodied intelligence robots demonstrated real-world applications in scenarios such as industrial manufacturing, cultural tourism, social governance, and special operations. The "Three New" selection set up Industrial Intelligent Agents as an independent category for the first time. Among the top ten achievements revealed, many innovations focused on perception, positioning, and measurement calibration, ranging from micro-voltage sensors in ultra-high voltage converter stations to calibration systems for humanoid robot dexterous hands.
Once AI truly enters the physical world, the challenges facing the industry change accordingly.
Over the past decade, the core metric for measuring IoT development has been the number of connections.
Based on trends and reports released at this expo, IoT Think Tank observed that profit margins for basic connectivity components are gradually under pressure, while industrial value is increasingly concentrating on high-performance hardware such as smart sensors and AI-specific chips, as well as application segments targeting industrial scenarios.
The center of gravity in industrial competition is shifting back from the connection layer to the perception layer.
Why the perception layer? Because information from the physical world must pass through the perception layer to enter digital systems.
What connections transmit, what decisions compute, and whether execution dares to act—all depend on the information acquired by perception.
This relationship was not prominent in the past because the ultimate users of sensor data were humans. But now, this "user" has become AI.
Data is now directly handed over to AI models and control loops. Algorithms lack the intuitive experience of veteran technicians and find it difficult to distinguish which sudden changes represent real physical variations. In the physical world, miscalculating a pressure value in pipeline scheduling or a one-millimeter deviation in a robotic arm can lead to accidents. When the recipient of data shifts from humans to AI, the requirements for sensors are restructured.
Previously, sensors only needed to solve the problem of "whether something could be detected." Now, they must answer two harder questions: First, "Effectiveness": Can filtering, compensation, calibration, and recognition be completed at the source to directly output usable state results? Second, "Reliability": Can the data achieve consistency, traceability, and metrological validity so that other systems dare to trust it directly?
How to make data both effective and reliable is precisely the key focus of the entire industry this year.
This year's "Three New" selection included "AI-based multi-sensor data fusion, self-calibration, and other algorithm models" in the scope of smart sensor solicitation for the first time. At the opening ceremony, organizations such as the GIIC Global Smart IoT Alliance, the LoRa Alliance, and the CSA Connectivity Standards Alliance jointly released the "Top 10 IoT Trend Insights for 2026," with "Perception Breaking Boundaries" listed as the leading main theme among three. From embodied perception and sensing-computing fusion to brain-computer interfaces and health perception, devices are evolving from "seeing" to "understanding."
If the past decade saw "connection" drive the scale of IoT, then in the AI era, "perception" will take the baton to realize IoT's value.
Exploring the Foundation: From Nearly CNY 100 Billion in Assets to 12 New Products, Hiding Three "Pivots" of Sensors
To discuss Wuxi's perception layer, one cannot bypass 2009. That year, the National Sensor Network Innovation Demonstration Zone—the only one in China—was approved, marking the start of "Perceiving China."
From 2009 to 2026, seventeen years of accumulation have allowed Wuxi to build substantial assets, with extremely high industrial clustering. In 2025, Wuxi's smart sensor industry revenue reached CNY 92.41 billion, a year-on-year increase of 10.3%. Among them, the Wuxi High-Tech District (including the Economic Development Zone), which started earliest, clustered 58.2% of the city's smart sensor enterprises in the first half of 2026, accounting for over 70% of the city's total operating income and output value.
Scale is just the background; more critical is where the industry is heading.
At this IoT Expo, 12 representative companies from Xinwu, Binhu, and Xishan each unveiled a new product, spanning inertia, acoustics, spectroscopy, force sensing, and optical chips.
A close look at these 12 new products reveals three clear pivots occurring in Wuxi's perception layer.
The first is "Computing Front-loading," where sensors no longer submit raw signals but finished products and conclusions.
Traditional sensors are responsible only for collecting analog signals or raw readings, with filtering, conversion, and recognition handled entirely by backend systems. However, in this year's new products, more than half of the companies moved data processing forward into the sensor itself.
At the chip level, Huajing Sensing launched high-performance MEMS microphones based on MEMS acoustic diaphragms + ASIC architecture for end-side devices like AI Pins and high-end wearables, directly outputting PDM digital signals to interface with SoCs, eliminating the ADC and wiring required for analog links. Memsensing Semiconductor introduced a single-chip absolute angle sensing chip for industrial motors and joint motors of embodied intelligence robots, based on AMR anisotropic magnetoresistance technology, integrating signal processing and DSP computation to achieve 0-360° full-angle detection, replacing traditional photoelectric encoders.
At the module and整机 (whole machine) level, industry algorithms are directly embedded in the body: Xunjie Guangyuan's MONET P online multi-channel Fourier transform spectrometer features built-in domestic AI edge computing and spectral early warning, with automatic model optimization, suitable for fields such as petroleum refining, chemicals, pharmaceuticals, semiconductors, food, and tobacco. Jianyan Smart IoT's AIoT stereoscopic perception monitoring system based on binocular stereo vision 3D reconstruction uses dual-path industrial cameras to extract features on-site, automatically calculating displacement magnitude and direction of the measured object. Zhongke Optoelectronics' particulate matter light quantum radar based on Mie scattering technology utilizes quantum-state near-infrared single-photon detection technology to rapidly invert optical parameters of atmospheric particulates, obtain concentration distribution information, and locate sudden emission sources in real time.
The second is "Calibration Entering the Factory," transforming site craftsmanship dependent on engineers into mass-producible factory configurations.
In the past, sensor reliability was hidden in engineers' on-site calibration and temperature drift compensation experience, making it difficult to replicate at scale. Now, Wuxi companies are moving reliability into the factory through processes and algorithms, addressing the most demanding data credibility issues for AI systems.
One path is completing complex calibration and AI compensation before shipment: Beiwai Sensing's embodied intelligence multi-sensor fusion pose navigation unit completes time "synchronization" and extrinsic parameter calibration of multi-source sensors before leaving the factory, reducing customers' secondary integration costs. Zhili Sensing's "high-precision six-axis force sensor adopting automotive-grade decoupling chips and end-side AI decoupling algorithms" uses MSG glass micro-fusion technology, employing AI decoupling algorithms to strip cross-interference between dimensions, combined with AI real-time temperature compensation to improve measurement reliability in wide-temperature environments.
Another path relies on semiconductor processes to ensure consistency: Lanrui Sensing integrates micro-structure humidity-sensitive capacitors, bandgap temperature-sensitive units, amplification, ADC, and digital calibration logic into a single CMOS chip, balancing accuracy, power consumption, and reliability of temperature and humidity sensing through mass-production consistency. Brilliance Electronics' passive fiber optic three-axis acceleration sensing system using silicon photonics reflection pulse encoding technology features a passive three-axis fiber optic sensing head that is resistant to electromagnetic interference and lightning strikes, with transmission distances exceeding 2km. It has been used in multiple wind farms to monitor turbine blade vibration posture and is also applicable to structural safety and equipment status monitoring in mines and bridges.
The deeper change is that micro/nano manufacturing capabilities are beginning to "cross boundaries," extending toward both the execution side of robots and the computing power side of AI.
Among the 12 new products, several have already gone beyond the traditional scope of "sensors," extending into two new directions.
One end extends to the embodied intelligence execution side: Lingzhang Robotics, a Beihang-affiliated company established only two years ago, launched a high-degree-of-freedom tendon-driven biomimetic dexterous hand, equipped with haptic sensors on fingers and palms, and a miniature camera integrated in the palm to achieve visual-tactile collaborative perception and feedback. Combined with Memsensing's single-chip absolute angle sensing chip for industrial and robot joint motors, Zhili Sensing's TJ60-50A six-axis force sensor, and Beiwai Sensing's embodied intelligence multi-sensor fusion pose navigation unit, the core components for humanoid robots from perception to action execution are now gathered here.
The other end extends to the AI computing power side: Tuoyi Technology, based on InP substrate technology, successfully overcame mode stability and heat dissipation challenges at high power, launching an InP-based high-stability mode heat-dissipation optimized CW-DFB laser chip for NPO/CPO400mW, providing autonomous and controllable core light sources for NGDC. Yuanmiao Systems' dual-axis quasi-static MEMS micromirror array chip for OCS optical circuit switches features mirror reflectivity >96% and step response <1ms, meeting the optical path scheduling needs of medium and large data centers. At the opening ceremony, the "Jiangsu Province 6G Communication Optical Bearing Technology Key Laboratory (Preparation)" was unveiled, led by Wuxi Deke Li Optoelectronic Technology Co., Ltd., jointly built with Nanjing University and Nanjing University of Science and Technology. With a planned cumulative investment of over CNY 120 million, its research directions cover ultra-high-speed optical bearing, air-space-ground core optical components, and communication-sensing-intelligent computing fusion.
If categorized by application scenario, these 12 new products concentrate on three high-barrier, long-cycle tracks: robot-level perception, energy and infrastructure safety-level perception, and computing power optical interconnect, perfectly echoing the "Smart Robots, Smart Terminals, Smart Computing Power, and Smart IoT" tracks of Wuxi's National Artificial Intelligence Industry Innovation Application Pilot Zone.
Building an Ecosystem: Constructing Public Platforms for "Long Slope Tracks" and Acting as "Ultra-Long-Term Partners"
At the opening ceremony of this IoT Expo, the National SME Public Service Demonstration Base (China IoT International Innovation Park) was unveiled. The China IoT International Innovation Park is Wuxi Micro-Nano Park, which successfully entered Jiangsu Province's sole list for creating National SME Public Service Demonstration Bases in 2026.
The award of this national-level designation to Micro-Nano Park hits the vital point of the sensor industry. Among the many niche areas of IoT, those most eager for "SME public services" are precisely smart sensors.
The reason is that sensors are naturally an industry dominated by SMEs.
With diverse categories, small individual market sizes, dispersed customers, and significant differences in technical routes, it is difficult to sustain the entire industry relying on a few giants, as is possible with terminal complete machines.
The 12 representative companies mentioned earlier are a microcosm. Four of them have been established for less than five years: Tuoyi Technology was founded in 2022, Lingzhang Robotics and Jianyan Smart IoT in 2024, and Zhili Sensing was just established in 2025. These companies have different backgrounds: Lingzhang Robotics spun off from Beihang University's Robotics Research Institute, Tuoyi Technology was formed by a global optoelectronics expert team, Jianyan Smart IoT was introduced from Shanghai, and Brilliance was created by merging the "Dual Academician Workstation" with the core R&D team of Juguang Technology.
For an SME sensor company, besides maintaining an R&D team, the three most money-burning aspects are: process lines and tape-outs, packaging and testing, and metrological certification.
A MEMS process line, a set of automotive-grade reliability tests, and an authoritative metrological certificate are too burdensome for a single SME to bear independently, yet they are ideal candidates for public services.
The role defined for the park as a demonstration base is to be an "ultra-long-term partner" for tech companies, providing full-chain services such as startup incubation, technical support, policy guidance, and financing/IPO assistance for IoT SMEs.
The term "ultra-long-term" is particularly apt when applied to sensor companies.
From tape-out to mass production, passing automotive certification, operating on the grid, or being written into a complete machine manufacturer's Bill of Materials (BOM), a sensor often takes several years to mature. During these years, the support needed by the enterprise constantly changes: space and incubation at the start, process lines and packaging/testing during prototyping, metrology and certification during customer introduction, and capital during expansion.
Micro-Nano Park holds 25 provincial-level or higher R&D institutions and an industrial fund cluster worth CNY 30 billion. It has built four high-energy industrial innovation platforms covering the entire chain of chip (device) design, manufacturing, and packaging/testing: the National Center for Advanced Packaging (NCAP China) - Wuxi Branch for Integrated Circuit Special Process and Packaging Testing, the Wuxi Wuyue Wuxin MEMS (Silicon-Based Optoelectronics) Advanced Perception R&D Center, the Wuxi Xichan Weixin Multinational Semiconductor IDM Group, and the National IoT Perception Equipment Industry Metrology and Testing Center. These provide full-chain and full-process services for enterprise growth, comprehensively solidifying the hard-core support for industrial innovation and development. As an incubator, the park solves whether a single company can "survive"; as a demonstration base, it aims to solve whether a group of companies can "grow and connect." The service target expands from single enterprises to industrial clusters, and the companionship period stretches from the initial few years of incubation to the entire process of enterprise growth.
This upgrade of Micro-Nano Park also signifies that Wuxi's approach to IoT industrial carriers is turning a new page.
In Conclusion
Viewing these 12 new products together with the upgrade of Micro-Nano Park, Wuxi's new positioning in the smart sensor track is already very clear.
Internally, SME sensor companies no longer need to face the financial pressure of tape-outs, packaging/testing, and certification alone. Relying on public service platforms, they can focus their energy on embedding algorithms, calibration, and compensation into the shipping process. Externally, Wuxi is growing into a complete perception and micro/nano supply chain base. Whether it is angle chips in robot joints, haptic sensors on dexterous hands, fiber optic sensing on turbine blades, or laser chips and MEMS micromirrors in data centers, downstream complete machine and computing power enterprises can achieve local matching here.
A deeper change is that the value of the entire IoT perception layer is being re-evaluated.
The business model of IoT is shifting from one-time hardware delivery to product-as-a-service and long-term operation, forming networked collaboration among chips, algorithms, complete machines, and scenario operators. When a sensor can output effective and reliable information at the source, it is no longer just an inconspicuous component in a machine but truly secures the most critical "data source" of the AI era.