SME  /  TECH & INNOVATION

By FedEx | July 17, 2026

​​Chinese startup Linkerbot has developed robotic hands that can play the piano, thread a needle, and tighten a screw in less than two seconds. Its co-founder, Su Yang, shares how the company is changing the game in dexterous manipulation.

Human evolutionary success has been shaped by two key traits: large brains and dexterous hands. While one drives complex thinking, the other enables precise physical execution. Today’s AI boom has propelled cognitive automation into the spotlight, as algorithms take on tasks that once required human intelligence. 

By contrast, progress in physical AI – such as humanoid robot hands designed to automate manual work – has advanced more gradually and remains less visible by comparison. One Chinese startup is working to close that gap. Founded in 2023, Linkerbot has developed a range of robotic hands designed for high-precision manipulation.

Take its flagship L30 model, which can tighten a loose screw in under two seconds and move with 0.2mm precision. Fingertip sensors allow it to detect pressure and regulate its grip, enabling it to assemble delicate parts without any fatigue. For manufacturing teams and production lines, this technology could enable a wide range of applications.

A longstanding barrier to the commercialization of robotic hands has been cost, with the most advanced models priced at hundreds of thousands of dollars. Linkerbot, however, has developed the capability to produce high-performance hands at scale. Its entry-level O6 model weighs just 370g but delivers a grip payload of 50kg, allowing it to hoist heavy loads. Most impressively, it’s 1/20th the price of competing products.

It’s no wonder that Linkerbot has quickly become a global market leader. The startup currently produces more than 1,000 robotic hands per month, cornering over 80% of the global high-degree-of-freedom (DoF) dexterous hand market. Its partners include the likes of tech giants Samsung and Siemens, along with research institutions such as Stanford, Cambridge, and MIT. It has also won a string of accolades, including a Special Mention at the 2026 FedEx Small Business Grant Contest.

How has Linkerbot grown from an early-stage startup into a leading player? We spoke to co-founder Su Yang to explore how the company is bridging the gap between experimental research and real-world applications in physical AI.

What inspired you to co-found Linkerbot?

Su Yang: The real trigger was not just seeing an opportunity but also recognizing a fundamental bottleneck in the field. In the early days of embodied AI, much of the focus was on the “brain” – models, algorithms, and intelligence. However, when we started bringing these systems into real-world environments, we realized that the true limitation lay in what we call the “last centimeter”: robots were not yet able to reliably touch, grasp, and manipulate objects.

In practice, robots often fall short not because they lack intelligence, but because they cannot interact effectively with the physical world. This led us to an important realization that dexterous manipulation is not merely a component challenge. It’s a foundational capability that shapes what robots can ultimately achieve.

That understanding is what inspired us to start Linkerbot. We chose to focus on building reliable, high-performance robotic hands – addressing one of the most difficult, yet essential, challenges in enabling robots to operate meaningfully in the real world.

How did you find your first customers?

In the early stages, our focus was less on actively seeking customers and more on understanding real-world problems where dexterous manipulation could create immediate value. Rather than relying on abstract brand positioning, we prioritized demonstrating what the technology could actually do in practice.

We invested significant effort into building tangible demonstrations, including complex manipulation tasks and live performance scenarios, so that potential users could clearly see our solution’s capabilities and limitations. This helped bridge the gap between concept and application.

At the same time, we focused on high-urgency use cases. Our early adopters were research institutions, humanoid robotics companies, and industrial partners who were already facing concrete challenges in manipulation. For them, this was not a “nice-to-have” technology, but something essential to progress.

In many cases, these early customers became close collaborators. Instead of a traditional vendor-client relationship, we worked together to define requirements, refine the product, and shape the direction of the solution.

Overall, our approach was grounded in a simple principle: in the early stages, success comes not from selling a product, but from solving a meaningful and clearly defined problem.

Can you share some of Linkerbot’s customer success stories?

Our work with customers has taken different forms, but they tend to follow a similar progression from early validation to real-world deployment. In research and advanced laboratory settings, we have helped teams move from limited or unreliable manipulation to stable, repeatable operation, enabling them to build more complete control and data pipelines.

As these capabilities mature, we see integration into broader robotic systems. Robotics companies incorporate our dexterous hands into their platforms, allowing them to move beyond demonstration scenarios and begin executing practical tasks in more structured environments.

In more advanced stages, this translates into commercial applications. Through our data collection systems and manipulation solutions, customers are able to train models more effectively and deploy systems at scale in real-world settings.

Across these different stages, one consistent outcome is that customers shift from showcasing what robots could potentially do to delivering systems that can perform useful work in practice.

What is one belief you hold that other business leaders might disagree with?

One belief I hold, which may differ from some perspectives in the industry, is that the pace and scale of AI commercialization are closely tied to advances in hardware capability. While AI is often viewed primarily as a software-driven field, the physical system plays a fundamental role in embodied intelligence.

Dexterous manipulation, in particular, serves as the interface between intelligence and the real world. Without reliable and scalable hardware, it becomes difficult to accumulate high-quality data, train models that generalize well, or deploy systems consistently in practical environments.

For this reason, we have chosen to focus on building foundational infrastructure, even though it is often more complex and less immediately visible than software progress. We believe that over time, these underlying capabilities will be critical in enabling AI systems to operate meaningfully and at scale in the physical world.

What is your greatest business failure that you treasure the most?

One of the most important lessons we learned came from a period when we placed too much emphasis on peak performance metrics, and not enough on overall system reliability. At the time, we were focused on achieving higher grip force, more complex motion demonstrations, and visually impressive results.

However, when these systems were applied in real-world scenarios, we began to see clear limitations. Stability was not always consistent, continuous operation was difficult to sustain, and deployment in customer environments proved more challenging than expected.

This experience prompted us to rethink how we define success. We came to realize that peak performance alone is not sufficient – what truly matters is whether a system can operate reliably and consistently over time.

Since then, we have shifted our priorities toward metrics that better reflect real-world usability, such as continuous operation capability, thermal stability, ease of maintenance, and task completion rates in practical environments. In retrospect, this experience was a valuable turning point that helped us move toward a more sustainable and application-focused approach.

What emerging trends do you foresee in the industry of embodied intelligence and physical AI?

From our perspective, the industry is undergoing several important shifts as embodied intelligence moves closer to real-world deployment. One of the most notable changes is a growing emphasis on real-world data. While simulation has played a critical role in early development, it’s becoming increasingly clear that large-scale, high-quality interaction data from physical environments is essential for building systems that can operate reliably outside the lab.

At the same time, the industry focus is evolving from demonstration to deployment. Earlier efforts often centered on showcasing what robots could potentially achieve. Today, the priority is on delivering systems that can perform useful tasks consistently in practical settings.

We’re also seeing a shift from standalone products to more integrated ecosystems. Future progress is likely to depend not only on individual components, but on how effectively data, hardware, algorithms, and real-world applications are brought together into a cohesive system.

Overall, these changes suggest that embodied AI is gradually transitioning from a primarily research-driven field to one increasingly shaped by real-world requirements and industrial applications.

How do you think physical AI will evolve in the future?

The development of physical AI is likely to progress in stages as systems become more capable and adaptable in real-world environments. In the near term, we expect to see continued improvements in controlled manipulation, where robots can reliably perform standardized tasks in structured settings.

Over time, these systems will become more flexible, enabling them to handle a wider range of objects and adapt to less predictable environments. This shift toward more generalized manipulation will be an important step in bringing robotic capabilities closer to the versatility we see in human interaction with the physical world.

Further ahead, we anticipate more autonomous and collaborative systems, where robots are able to understand high-level objectives, make decisions, and work alongside humans or other machines in a coordinated way.

Across all these stages, one principle remains central: the long-term value of AI lies not only in its ability to process information, but also in its ability to interact with the physical world. In that sense, progress in areas such as manipulation, touch, and real-world data will continue to play a foundational role in shaping the future of the field.