A robot can recognize an egg on a table, but it can't necessarily pick it up.
This is the real gap faced by physical intelligence today: visual and decision-making models are improving rapidly, and “hand” abilities, particularly the perception of touch, pressure, and slippage, are still a shortcoming.
The Zhitong Finance App learned that Ledong Robotics (01236) issued a voluntary announcement on the evening of October 11. The company recently reached a strategic cooperation with Huazhong University of Science and Technology to conduct joint research and development on flexible 3D tactile sensors and array sensing systems for embodied robots. A robotics company that started with “eyes” (lidar, visual perception) is betting on “skin” for the next step.
The bottleneck of physical intelligence is shifting from “seeing” to “doing”
Over the past two years, large models have rapidly evolved the “brain” of robots, and visual perception solutions have also matured. However, in order for robots to enter the factory and home from the showroom, they must complete real physical interactions such as grasping, assembly, and collaboration. At this point, vision is not enough: blocked objects, fragile items, parts with smooth surfaces, and the strength required for precise control are all beyond the scope that can be judged by sight.
Humans complete a grip with real-time feedback of pressure, cut, and slip from the fingertips. The robot lacks this level of feedback; it can only “blindly catch it with its eyes”; either it can't grab steadily or it's too hard. As a result, the industry gradually reached a consensus. Vision determines what a robot “knows,” and touch determines what a robot “can do.” For dexterous hands and humanoid robots to achieve everything from demonstration to practical use, tactile sense is a part that cannot be circumvented.
Why “3D”? This is a watershed moment for tactile perception
Tactile sensing is not a new concept, but most existing solutions can only sense “multiple pressure”, that is, normal forces perpendicular to the contact surface. However, in actual operation, the more critical information is hidden in two other directions: tangential forces in the two directions before and after contact, determine whether the object is falling and whether the finger is being dragged. Only by receiving force in these three directions at the same time can the robot determine whether the grip is steady or unstable, and then adjust the strength and posture.
This is also the reason why 3D haptics has become the focus of industry research. It is the data foundation for dexterous grasping, fine assembly, and human-robot collaboration. But it's also very difficult. The sensing unit must not only be soft enough to fit a curved surface, but also withstand repeated contact. Force signals in the three directions must also be accurately decoupled by an algorithm. The array density is high, and the difficulty of simultaneous acquisition and packaging also increases exponentially. Problems occur in any of the materials, microstructures, packaging, and algorithms, and there is no way to talk about mass production.
Demand has been verified, standards have not yet been formed, and the tactile circuit is in the window
Judging from the industrial process, 3D haptics is currently in a typical early stage. Demand has been verified, but technology routes and industry standards have not yet converged. The data collection side has taken the lead in generating demand, but it will still take time for the execution side to scale up.
What is more noteworthy is that mainstream embodied intelligence models are currently not connected to tactile modes on a large scale. Once the head model begins to systematically introduce tactile data, hardware supply capacity will quickly become a bottleneck. Therefore, the real valuable time difference appears before the big model is connected. Whoever can build, stabilize, and quantify the sensor first has a pioneer in data accumulation and engineering experience.
According to the Insight Consulting Report, other sensing technologies, including hearing, touch, and force, are expected to maintain a compound annual growth rate of 20.8% from 2024 to 2029, and the global market size will reach 34 billion yuan in 2029. The track is still small, but it is on the eve of the explosion.
A reasonable extension of a “perception company”
Seen in this context, it is not surprising that Ledong robots chose three-dimensional haptics. The company has long accumulated sensing software and hardware around lidar, SLAM/VSLAM, and AI scene recognition, and has formed a two-wheel drive model of “visual perception product+complete robot”, and has completed commercial verification in scenarios such as mowing robots. Extending from “seeing” to “touching” is a step in understanding the company's multi-modal expansion.
According to the Ledong Robotics announcement, the two sides will rely on the National Key Laboratory of Intelligent Manufacturing Equipment and Technology of Huazhong University of Science and Technology to develop flexible 3D tactile sensor devices and array systems around the tactile sensing requirements of physical robots, including three directions: developing flexible 3D tactile sensors and high-density sensing arrays to improve sensing ability and long-term stability under complex operating conditions; developing supporting signal acquisition hardware, 3D force decoupling and tactile imaging algorithms to form an overall solution from sensor components to array systems; and in addition to dexterous hands, robotics, humanoid robots, collaborative robots, etc. Verify the ability to land the scene on the carrier.
Professor Li Kan, who led this collaboration, is currently a doctoral supervisor at Huazhong University of Science and Technology and winner of the National Natural Science Foundation of China Outstanding Youth Science Foundation Project (Overseas). He has long been committed to research on three-dimensional flexible structures and stretchable electronic devices. He has published many papers as the first author or co-first author in journals such as Nature, Nature Communications, Science Advances, and PNAS.
Judging from the cooperative structure, this is a typical division of labor between industry, education and research. Three-dimensional flexible tactile sensing spans flexible electronics, special materials, precision packaging, and force decoupling algorithms. The threshold for basic research is high, and requires university teams to continue to tackle cutting-edge scientific research. However, in order for sensors to actually become products, they also need to be repeatedly verified throughout the entire scene. This is the value of the participation of Ledong robots. What this type of cooperation should address is the “valley of death” between principle prototypes and mass-producible products.
Looking at the business implementation path, 3D tactile sensing can not only be used as an independent sensing product line for customers such as dexterous hands, robotic arms, humanoid robots, collaborative robots, etc., but also take the lead in verifying implementation on Ledong Robotics' own complete machine products, forming a differentiated hardware advantage.
As far as the capital market is concerned, this is an extension of the Ledong robot's “perception infrastructure” narrative into multiple modes. As the robot industry moves from technology verification to large-scale application, upstream sensing components are often the first to benefit from category expansion and shipment volume. The company not only provides sensing products to different types of robots, but also promotes technology implementation through its own machines. The business structure has a certain degree of collaborative flexibility.
However, it still takes time for 3D haptics to move from laboratory to mass production, and R&D progress and commercialization pace still need to be continuously tracked. At the beginning of September this year, the company was included in the Hong Kong Stock Connect list. At one point, the stock price rose from HK$23.5 to HK$31.68 in October, the biggest increase of 34.8%. Investors can pay attention to the company's subsequent R&D progress and order fulfillment.