Self-developed fingertip-level bionic visual-tactile sensors, "Yimu Technology" aims to equip robots with "eyes that can see tactile sensations"
With the rapid development of fields such as humanoid robots, industrial automation, and household intelligent devices, "perceptual ability" has become a core indicator for measuring the intelligence level of robots. Currently, perceptual technologies such as vision and hearing are relatively mature, but touch, as the "last mile" for robots to interact with the physical world, has long been restricted by technological bottlenecks.
Traditional visual - tactile solutions generally have pain points such as large volume, obvious temperature drift, blind spots in perception, and dependence on imported materials. They are difficult to meet the application requirements under refined operations and complex working conditions, and have become a key obstacle restricting the upgrade of robots to "dexterous operation" and "autonomous decision - making".
Against this background, Yimu Technology, which has been deeply involved in the field of robot perception for ten years, recently released the world's thinnest bionic visual - tactile sensor, opening up new imagination space for the commercial implementation of "tactile intelligence" in the industry.
Founded in 2016, Yimu Technology is a company focusing on perception - driven AI computing. In terms of business layout, Yimu Technology has formed three core sectors: instrument intelligence, electrical appliance intelligence, and embodied intelligence.
Among them, the "instrument intelligence" business centered on the chipization of molecular spectrometers has achieved mature commercialization, with an annual revenue of hundreds of millions of yuan and continuous profitability. It is mainly used in scenarios such as water quality testing and laboratory analysis, successfully realizing the domestic substitution of European and American products. The "electrical appliance intelligence" business targeting the AI transformation of household robots and home appliances is in the early stage of cooperation with major customers and is expected to achieve large - scale growth in the next one or two years. The "embodied intelligence" business centered on visual - tactile sensors has won a large number of orders and is expected to experience explosive growth next year.
The instrument intelligence business is currently still the main revenue source, while embodied intelligence is regarded as the core engine for future growth. The idea of entering the embodied intelligence business originated at the end of 2023.
Before officially launching the project, the team spent a lot of time researching the industry's needs, from dexterous hand manufacturers, robot body enterprises to gripper suppliers. Finally, they found that "excessive volume" was the most concentrated pain point reported by customers. "The thickness of traditional visual - tactile solutions generally exceeds 20 millimeters. Not to mention integrating them into the hands of small robots, even when adapting to mainstream dexterous hands, they will occupy the operating space and affect the flexibility of movements, making them unusable in many scenarios," said Li Zhiqiang, the founder of Yimu Technology.
Therefore, developing an extremely thin and lightweight visual sensor suitable for all scenarios became the R & D goal of Yimu Technology. For this purpose, the team comprehensively optimized from geometric optical design, model calculation to chipization technology and presented the results at the IROS Global Robotics Conference in October this year. It is reported that its bionic visual - tactile sensor uses a contact surface in the shape of a human finger pad, with its size and thickness approaching those of a human fingertip, and the thickness is only half of that of similar products in the industry.
To address the pain points of traditional solutions such as temperature drift at different temperatures and perception blind spots, the team incorporated self - developed perception algorithms to achieve a design with no blind spots in full calculation and reached the IP65 protection level, ensuring stability and accuracy under complex working conditions.
In terms of perception performance, the sensor has a micron - level deformation calculation accuracy, a force resolution of 0.005N, and a maximum output frame rate of 120fps. With this performance combination, the robot can detect extremely slight pressure changes and output data in real - time at a high rate, providing timely and accurate tactile feedback for refined operations.
In terms of engineering reliability, by optimizing the process of wear - resistant soft elastomers and Marker marking points, Yimu Technology has broken through the limits of materials science, enabling the product's mechanical properties to pass long - term durability tests. This means that the sensor can not only work stably in the laboratory environment but also withstand the severe tests of various complex practical application scenarios.
The breakthrough in materials science is even more crucial. Different from the current situation where most enterprises in the industry rely on imported materials from Europe and the United States, Yimu Technology has established a professional team to self - develop flexible electronic skin materials. This not only breaks the "choke point" risk but also achieves stable performance under millions of presses and tens of thousands of lateral shear forces, with significantly improved sensitivity and consistency. This means that the sensor can not only work stably in the laboratory environment but also withstand the severe tests of various complex practical application scenarios.
In addition, Yimu Technology has adopted a self - developed full - stack simulation workflow in optical design, narrowing the Sim - to - Real gap. This technology enables the sensor to complete a large number of tests and optimizations in a virtual environment, significantly accelerating the product iteration process and effectively reducing development costs.
"The birth of this sensor is the result of the collaborative efforts of our interdisciplinary team. From materials to chips to algorithms, self - development across the entire chain ensures the core competitiveness of the product," emphasized Li Zhiqiang.
Why have the long - standing industry pain points not been solved for a long time? Li Zhiqiang gave two core reasons: First, visual - tactile technology involves multi - disciplinary intersections and requires the integration of capabilities in fields such as materials, optics, and chips. Most peer teams have a short establishment time and a single discipline, making it difficult to cover all aspects comprehensively. Second, the previous technical solutions were immature, resulting in the market demand not being activated, and customers had insufficient awareness of the application value of "tactile perception".
It is reported that after the product was launched, Tesla, leading domestic robot enterprises, and core suppliers have all extended olive branches and signed cooperation contracts. The product has now entered the stage of commercial application.
Li Zhiqiang introduced that installing this sensor is equivalent to equipping the robot with "eyes that can see touch". Its core technology lies in capturing the subtle deformation of the elastic material when it contacts an object through the built - in camera, obtaining a sequence of high - definition "tactile photos", and calculating rich tactile signals through AI.
In addition, Yimu Technology's bionic visual - tactile sensor can simultaneously "feel" the softness and hardness of an object, its surface texture, and even the sliding trend. This high - fidelity tactile information enables the robot to more accurately identify the characteristics of the object and complete various delicate operations like a human being.
"To break the foreign monopoly, we must make breakthroughs in underlying technologies rather than simply copying," Li Zhiqiang emphasized when talking about the experience of domestic substitution. Taking the chipization of molecular spectrometers as an example, the team abandoned the technical route of traditional large - scale instruments in Europe and the United States. Based on the scientific research achievements accumulated by the founder during his doctoral studies, they used chipization technology to significantly reduce the cost and volume and incorporated AI algorithms, compressing the entire process of sampling, analysis, and report generation from the traditional one or two days to two or three seconds.
This idea of "underlying innovation + AI integration" is also applied to visual - tactile sensors. The team dissected the technical pain points of traditional solutions and carried out underlying research across the entire chain from optical chips, materials science to algorithms, rather than simply imitating, and finally achieved performance surpassing.
For future planning, Yimu Technology will continue to focus on the field of embodied intelligence. In the short term, in 2026, it plans to release an improved VTLA model integrated with a tactile alignment algorithm to fill the gap in the current large - scale models in the field of tactile perception. At the same time, it will promote the iteration of the next - generation visual - tactile sensor, aiming to achieve "cliff - leading" indicators such as a response time of 2 milliseconds and a thickness of less than 2 millimeters.
In the long term, in the next 3 - 5 years, it will advance along two lines: "exploring the upper limit of touch - based embodied intelligence technology" and "improving industrial implementation". On the one hand, it will pursue a multiple - level improvement in sensor performance. On the other hand, it will reduce costs through full - stack self - development, automated production, and bulk procurement to achieve large - scale implementation.
Facing the increasingly fierce industry competition, Li Zhiqiang believes that the key to maintaining leadership lies in the dual construction of "strategic determination" and "systematic capabilities". Strategically, the company focuses on the field of perception and related algorithms and does not involve in the links of robot bodies and grippers. It builds a good ecosystem through open cooperation. "Some peer companies blindly expand their business boundaries and may even compete with customers, while we focus on the core field and can gain more trust from partners," Li added.
In terms of systematic capabilities, Yimu Technology has built a complete process from market insight, basic research, product R & D to industrial mass production. The team took nearly ten years to build and has gathered doctoral talents from top global universities such as CMU, Harvard, and Tsinghua, covering multiple fields such as materials science, optics, chips, and algorithms.
When talking about the current shortcomings, Li Zhiqiang frankly said that although the system is perfect, the shortage of human resources is still the primary challenge, especially in the context of the booming embodied intelligence, where the competition for high - end talents is fierce. Secondly, there are still gaps in the authenticity of tactile simulation and the full - link capabilities of AI large - scale model closed - loop control in the industry, which require continuous investment to break through. In addition, capacity expansion and operation optimization are also directions that need to be improved in the future.