From “Robot Eyes” to Physical AI Data Portals: RoboSense’s Four Major WAIC Collaborations Position It in the Embodied Intelligence Supply Chain
From July 17 to 21, 2026, during the World Artificial Intelligence Conference (WAIC) in Shanghai, RoboSense (02498.HK) announced strategic partnerships with four companies. The partners cover embodied foundation model development, real-world data collection, data-evaluation-deployment feedback systems, and expansion into the Middle Eastern market.
According to the company, RoboSense will leverage its proprietary spatial perception technology as a fulcrum to deeply integrate into the physical AI data industry chain.
The “Bottleneck” of Physical AI: Lack of Data
For embodied intelligence to enter the physical world, the bottleneck is neither computing power nor algorithms. Digital AI large language models are built upon 30 years of data accumulation from the global internet, trained on trillions of tokens, while the real interactive data available for embodied models is less than one ten-thousandth of that.
According to a research report by Guosen Securities, to achieve usable embodied intelligence, at least 10 million hours of multimodal data are required; factors like multi-scenario and long-tail distribution mean actual demand far exceeds this number. Companies like MiFeng have already raised their 2030 target data collection volume to 10 billion hours.
The data required for physical AI is fundamentally different from the text and image data of the internet era. Xie Tiandi, Marketing Director of RoboSense, stated in an interview: “Traditional video data is actually relatively coarse for embodied large models. It requires a more complete spatial data structure and more precise depth information.”
Pure visual sensors are susceptible to interference from strong light, occlusion, rain, and snow, and have inherent errors in ranging and spatial structure measurement. The physical rules constructed in simulation environments differ significantly from real-world operating conditions. The “bottleneck” of physical AI is precisely high-quality 3D spatial data.
Strategic Positioning: From “Selling Sensors” to “Data Gateway”
The four strategic agreements signed by RoboSense at WAIC precisely correspond to four key nodes in the physical AI data value chain.
Strategic partnership with Guanglun Intelligence (a physical AI data and evaluation infrastructure service provider): Starting from model training and evaluation needs, the two parties will reverse-define collection hardware and data specifications. Guanglun Intelligence predicts that in 2026, data and evaluation demand in the embodied intelligence field will see exponential growth, with the overall scale reaching 100 to 1000 times that of last year.
Strategic partnership with Jianzhi Robotics (an embodied intelligence data infrastructure enterprise): The two parties will engage in deep collaboration around real-world multi-view Ego data collection and other businesses to jointly build a high-quality, multimodal 3D physical world data system. Jianzhi Robotics’ self-built data production lines have covered over 10,000 real-world scenarios, accumulating millions of hours of multimodal human operational data.
Strategic partnership with Zhizai Wujie (an embodied general foundation model developer): To connect the “perception-cognition” technology loop. Zhizai Wujie’s Being-H/Being-M general large models can already be deployed in real-time on edge devices.
Strategic partnership with Origen (a UAE-based AI-native technology company): To expand the embodied intelligence ecosystem in the Middle East. Origen’s smart home brand DOMIA and spatial intelligence platform StellaWare are already widely used in the UAE market.
Yang Xiansheng, Vice President of RoboSense, stated at ICRA 2026: “Currently, robots lack high-quality spatial data, which is why our sensors have emerged.” Yang Haibo, President of Guanglun Intelligence, pointed out in his WAIC speech: “We hope to start with model training and evaluation needs, reverse-define collection hardware and data specifications, so that new modal capabilities can truly enter the same continuous learning infrastructure.”
The common direction of the four agreements is: RoboSense is not just selling sensors, but using its proprietary hardware as a fulcrum to deeply integrate into key links of the physical AI data industry chain.
Three Moats: Chips, AI, Mass Production
RoboSense’s ability to position itself as a “data gateway” relies on the deep integration of three core capabilities.
First, root technology at the chip level. The company’s self-developed “Peacock” SPAD-SoC chip is currently the industry’s only mass-producible highest-specification dToF area array chip. The second-generation fully solid-state digital lidar E2, built on this chip, achieves a wider field of view in a more compact body, with maximum accuracy improved to 3 times that of the previous generation and a point frequency reaching the million level. The “Peacock” chip is “mass-produced upon release” and will be shipped in large quantities in the third quarter of 2026. The company is driving the upgrade of robot spatial perception from “device stacking” to “chip-defined.”
Second, an AI-driven data closed loop. RoboSense incorporates the demand for high-quality spatial data from model training into chip and product design from the outset. Robots equipped with its perception products can synchronously generate computable and learnable data assets during movement and operation.
Third, automotive-grade mass production capabilities. The company has established an automotive-grade system covering R&D, testing and verification, supply chain, manufacturing, and delivery. As of March 2026, it has secured design wins for 177 vehicle models from 36 automakers and Tier 1 suppliers. The high-standard engineering capabilities of automotive-grade mass production constitute a composite barrier that is difficult to overcome quickly through single-point technology.
In the first half of 2026, the company’s sales of 3D lidar for the robotics field reached 282,600 units, a year-on-year increase of 510%. According to GGII data, the company has ranked first globally in 3D lidar shipments for multiple consecutive quarters, leading in five major sub-markets: lawn mowing robots, unmanned delivery, humanoid robots, embodied intelligence, and commercial cleaning robots.
One Hardware, Two-End Markets
RoboSense has a unique structural characteristic in the industry chain: the same underlying hardware simultaneously serves the “front-end” and “back-end” of the physical AI industry chain.
The back-end is the collection of large amounts of high-quality 3D spatial data required for large model training. Physical AI large models require massive real-world 3D perception data, and RoboSense’s spatial perception hardware is the “production equipment” for this type of data. The partnerships with Jianzhi Robotics and Guanglun Intelligence target this back-end market.
The front-end is the real-time perception needs of various types of embodied robots. From humanoid robots and quadruped robots to lawn mowing robots and commercial cleaning robots, every robot entering the physical world needs to “understand” the 3D environment. The Zhiyuan humanoid robot equipped with RoboSense’s fully solid-state digital lidar completed demonstrations such as dance performances and voice interactions at the WAIC site.
The “data isomorphism” between the back-end and front-end means that the same series of hardware serves as both the “eyes” for robot operational perception and the “data collector” for physical AI model training. Both ends share the same technology platform and the same series of products. This network effect of the data closed loop is the core characteristic that distinguishes RoboSense from pure hardware suppliers and pure data collection companies.
From “Lidar Company” to “Physical AI Infrastructure Company”
The market’s perception of RoboSense still remains within the framework of a “lidar supplier.” However, the series of actions at WAIC 2026 indicate that the company’s strategic positioning has undergone a fundamental leap.
In the first quarter of 2026, RoboSense had over 3,400 global robot customers, with lidar sales in the robotics field reaching 185,500 units, a year-on-year increase of 1458.8%. Its share in total sales surged from 11% in the same period last year to 56%, surpassing the ADAS automotive business for the first time to become the company’s largest revenue source, marking RoboSense’s entry into a new phase.
Future Markets estimates that the global physical AI market will grow from $383 billion in 2026 to $3.26 trillion by 2040. In this incremental market, data infrastructure construction is expected to be the first to see an investment peak.
RoboSense uses its self-developed 3D spatial detection chip as the foundation, builds a spatial perception product matrix covering various robot forms, connects robot bodies with embodied models, and transforms high-quality spatial data into the fundamental capability of robots. From the “eyes of robots” to the data gateway for physical AI, RoboSense is defining its industrial coordinates in the era of embodied intelligence.
Source: Bot Telegraph China