On September 7, embodied intelligence company Zhixing Embodied officially unveiled Imprint X, a vision-tactile sensor designed for robotic manipulation. Unlike existing tactile-sensor approaches on the market, Imprint X uses an event-driven, neuromorphic vision-tactile sensing architecture to create a commercially viable robotic sense of touch.

By combining innovations in underlying sensing, algorithms, materials, and structural design, Imprint X breaks through the performance constraints of conventional vision-tactile sensors. It delivers a frame rate of more than 1,000 Hz and latency as low as 1 ms, bringing high-precision, fast-response, low-latency, low-power vision-tactile sensing to robot fingertips.

Zhixing Embodied has thus become the first company to take this type of product out of the lab and into the market. The company has reportedly begun real-world validation on production lines at multiple leading 3C manufacturers and has reached a product partnership with a leading domestic dexterous-hand maker.

(Imprint X vision-tactile sensor)

Fine Robotic Manipulation Calls for Tactile Sensing

As embodied intelligence moves from simply being able to act to being able to do work, robots are entering a growing range of fine-manipulation scenarios.

With vision, robots can often perform basic tasks in simple grasping operations. But when the task involves precision insertion, flexible assembly, or handling fragile objects, robots need more than sight: they need a sense of touch.

In precision assembly, for example, humans continuously combine vision and touch to adjust their movements in real time while assembling mobile-phone components. Robots without tactile sensing tend to force parts together: apply too much force and an interface may jam; misalign the angle and a component may be damaged.

When a robot end effector is equipped with an Imprint X vision-tactile sensor, it can detect force, deformation, slip, and other signals. At the instant of contact, a large model can continuously adjust the robot's movement, determine whether a tiny connector is aligned or jammed, and promptly correct the angle and force when needed, or back off and realign.

For robots to move beyond a one-off demonstration to stable, repeatable work in real environments, they urgently need this kind of perception loop. Touch is becoming a foundational capability for fine robotic manipulation.

Breaking the Frame-Rate Barrier With Event Cameras That Sense Change as It Happens

Tactile sensors follow several technological approaches. Vision-tactile sensors can detect deformation on a contact surface, providing richer information than force sensing alone.

Zhixing Embodied chose an event-driven, neuromorphic sensing approach for Imprint X. Combining innovations in underlying sensing, algorithms, materials, and structure, Imprint X overcomes critical bottlenecks in response speed, frame rate, and power consumption that have constrained existing vision-tactile sensors, bringing high-precision, fast-response, low-latency, low-power vision-tactile sensing to robot fingertips.

Conventional vision-tactile systems typically use frame-based cameras to capture contact deformation frame by frame. During rapid contact or dynamic slip, their limited frame rates severely restrict their ability to capture change.

Zhixing Embodied chose a different route from the outset. Imprint X is built around an event camera, which responds primarily to changes in a scene rather than capturing full images, outputting incremental data only when pixels change.

This enables Imprint X to achieve latency as low as 1 ms and frame rates above 1,000 Hz, allowing robots to receive tactile changes during high-speed contact within milliseconds.

Because it does not need to continuously output large volumes of repetitive image data, the sensor maintains a low data rate even during high-speed sensing. The system requires no additional integrated computing power, while bandwidth use and power consumption are substantially reduced.

For robots, detecting contact without understanding what is happening during that contact still makes high-speed touch difficult to translate into manipulation capability. Imprint X, however, captures rich multidimensional tactile information.

Imprint X uses a proprietary contact material embedded with an array of marker points. When contact deforms the material, the event camera captures changes in those markers, while proprietary algorithms modeled on event-data streams reconstruct the deformation and force state of the contact area. Imprint X can do more than detect whether contact has occurred: it continuously outputs a 2D displacement field, a 3D tactile deformation point cloud, and tactile features including distributed force and slip.

For engineering deployment, Imprint X is compact and easy to integrate into fingertips or grippers. Its low-data-rate, low-power design reduces integration barriers, while its enclosed structure reduces interference from ambient light, dust, moisture, and other environmental factors. Its contact material uses a modular quick-swap design, allowing worn components to be replaced individually without replacing the entire sensor.

Full-Stack Strategy: From Vision-Tactile Sensing to Physical AI

In Zhixing Embodied's view, sensors are only the starting point for giving robots a sense of touch. As a full-stack company focused on embodied intelligence and physical AI, its goal is not merely to build a vision-tactile sensor, but to turn touch into manipulation capabilities that robots can learn, understand, and execute.

To achieve that goal, Zhixing Embodied is building a closed loop spanning devices, data, and models. On one side, it uses hardware such as Imprint X to collect tactile information from the physical world; on the other, it is collecting human-operation data and translating experience in fine human manipulation into data and policies for robots. At the same time, the company is developing embodied intelligence large models that incorporate the tactile modality. Within this loop, sensors determine whether robots can perceive, data determines whether they can learn, and models determine their ultimate ability to translate that into action.

Zhixing Embodied is led by serial entrepreneur Fei Bibo. Its chief scientist, Cui Shaowei, comes from a key laboratory at the Chinese Academy of Sciences' Institute of Automation, while CTO Hu Xin has extensive experience in big-data processing and search-engine software architecture. Its core team spans robotic touch, fine operations, simulation systems, representation models, and large models.

Imprint X has now entered real-world validation in precision-manipulation scenarios on the production lines of several leading 3C manufacturers, while Zhixing Embodied has also reached a product partnership with a leading dexterous-hand manufacturer. Within the industry, it has become the first company to bring products based on this type of solution from the lab to market.

Zhixing Embodied aims to make touch a foundational capability for robots, enabling them to sense every interaction with the world and ultimately unlock the physical world's “last millimeter.”

[This article is published with authorization and does not represent the views of Zero2IPO. This platform provides information storage services only.] [Disclaimer]: This article does not constitute investment advice. Markets involve risk, and investment decisions should be made with caution. For any questions, please contact Zero2IPO at editor@zero2ipo.com.cn.