At the 2026 World Robot Expo (WRC), held from August 19-23 at the Beijing Etrong International Convention and Exhibition Center, a special forum on "AI Large Models Enabling New Paradigms for Robotics and Embodied Intelligence Industries" was held alongside the main event. Zhu Honglei, CTO of Wanna Robotics, participated in the roundtable discussion titled "Scenario Application Innovation and Industrial Chain Collaboration," sharing insights from the frontlines of dexterous hand development and deployment.
Zhu explained that Wanna Robotics has built its foundation on micro servo electric cylinders since its inception, using these components to construct their dexterous hand products, which are then deployed in real-world scenarios to address genuine, high-demand needs. Through iterative feedback loops from actual applications, the company continuously refines both their dexterous hands and micro servo cylinder offerings. The overarching goal, he noted, is to bring these products into labor-intensive, high-strain environments where they can replace repetitive and physically demanding manual work.
When asked about the company's recent new product launch and his extensive background in robot motion control, Zhu shared practical lessons from their dexterous hand deployment, including a 7x24-hour uninterrupted sorting operation during the 618 shopping festival. He revealed that the company originally sought to purchase suitable micro servo electric cylinders for their dexterous hands, but finding no products on the market that met their requirements, they decided to develop their own. This led to a careful balancing act between cost, performance, and reliability across both the cylinders and the dexterous hands themselves.
The design process was driven by real industrial scenarios that require three-shift operations or involve simple yet highly repetitive and intense labor. These scenarios shaped clear preferences for the cylinder control capabilities, prompting questions about force output strength, whether speed could be traded off, and what levels of durability and velocity were needed to meet production line rhythms. When the dexterous hands were deployed in their own pilot scenarios during the 618 promotion, the environment pushed the technology further. Soft packages proved particularly challenging to grasp reliably, requiring the integration of models that propose grasping poses, combined with sensor data and fingertip tactile feedback, alongside visual fusion across multiple dimensions.
Zhu emphasized that effective use of the hand demands both strong model capabilities and a hand that faithfully executes model commands with efficiency. This splits into two major areas: on the model side, determining whether and how to grasp, breaking down detailed movement sequences, timing, and multi-axis coordination. Through iterative upgrades, Wanna Robotics has embedded these relatively simple but valuable algorithms directly into the dexterous hand itself, exposing them through separate APIs. This integration simplifies model usage by incorporating tactile sensing and pose control, allowing simple commands to achieve effective grasping and higher success rates.
Real-world deployment also revealed communication bottlenecks. The 485 communication protocol originally used proved insufficient for high-degree-of-freedom hands with rich tactile feedback. Initially, the hand and arm required external cabling, which was prone to snagging or interference in practical use. In their iterative upgrades, Wanna Robotics worked with collaborative robot arm manufacturers to integrate the hand internally, resulting in cleaner external setups and improved communication speed and stability. With internal stability achieved, all interfaces route from the robotic arm to the central controller, eliminating external wiring issues. They also introduced EtherCAT communication for high-degree-of-freedom systems, enhancing data collection frequency, dimensionality, and real-time performance. This entire journey, from micro servo cylinders to dexterous hands to real-world deployment, creates a feedback cycle where scenario insights guide product improvements, ultimately boosting performance, stability, and operational success rates.
Addressing the challenges of moving from simulated actions to stable warehouse operations, Zhu acknowledged that initial expectations were far simpler than reality. Starting from simulation, grasping seemed like a well-defined action, but the complexity of the real world far exceeded simulated environments. Many issues only surface when robots are placed in actual scenarios. For instance, grasping soft packages initially followed ideal geometric assumptions, such as maintaining a 1mm gap between fingers for stable pickup. However, in real settings, soft packages deform, and combined with inevitable minor execution deviations, that precise 1mm gap produced wildly different results across packages. Slightly wider gaps led to unstable grips, while narrower gaps caused excessive compression. This realization drove home that real-world grasping cannot rely solely on preset positions and fixed actions; robots must perceive objects and adjust dynamically based on feedback.
Through continuous iteration based on field feedback, Wanna Robotics integrated tactile sensing into the main control system. After the dexterous hand contacts a package, the controller evaluates contact status via tactile feedback and autonomously calls upon the actuator module's angle control capabilities to dynamically adjust finger and joint positions, keeping grasps within a reasonable range. This approach improves grasp stability while minimizing product damage, enhancing overall quality rates. Real warehouse environments also present non-standard objects, with varying dimensions, shapes, and hardness levels, each requiring different hand poses, joint configurations, and control strategies. The team continuously tests, gathers feedback, and iterates in real scenarios, distilling experiences into algorithms and control systems.
Zhu described this as an evolutionary process for dexterous hand capabilities: starting from merely executing grasping commands, progressing to incorporating tactile sensing and algorithms for autonomous adjustment based on contact states, and moving toward making independent grasping decisions for different package types. For Wanna Robotics, the value of real-world scenarios extends beyond verifying whether a robot can grasp; it lies in integrating perception, decision-making, and execution through repeated real tasks, enabling the dexterous hand to shift from "grasping as instructed" to "deciding how to grasp based on the situation," ultimately improving overall success rates and stable operational capability.
On the pressure of customer delivery expectations, particularly in the highly efficient logistics industry with stringent real-time requirements, Zhu acknowledged that pressure certainly exists. However, he noted that logistics scenarios differ fundamentally from high-precision assembly lines like automotive manufacturing in how robots are evaluated. Logistics focuses less on absolute zero-error for any single action and more on whether a robot can operate stably over extended periods within real business environments, continuously completing tasks and ultimately improving overall workflow efficiency. In front-warehouse sorting, for example, the site operates as a systematic process with task scheduling and exception handling mechanisms ensuring overall fulfillment. The critical factors for robots are maintaining stable grasping, recognition, and placement capabilities during high-frequency, continuous operations while coordinating seamlessly with WMS, robotic arms, and vision systems.
Therefore, Zhu emphasized that the core focus during on-site delivery is not merely single-attempt success rates but whether the entire system can run stably and sustainably under real business pressure, creating genuine efficiency value. This is precisely why Wanna Robotics consistently stresses the importance of validating products in real-world scenarios.
Comments