The World Humanoid Robot Games began in China on Saturday (22) and will continue until Wednesday (26). This event represents the pinnacle of the sector, where China has invested significantly over the last decade.
Technology and robotics companies use this event to showcase their products. However, for the 2026 edition, they face two crucial challenges: demonstrating that their robots can perform delicate movements with precision and proving their utility and productivity in a real environment.
Over the past ten years, China has transformed robotics competitions into strategic platforms for businesses. Between 2014 and 2015, Xi Jinping, then president of the Asian country, designated robotics as a central element of Chinese manufacturing.
Since then, the focus of these robotics gatherings has shifted from mere academic exhibitions to the execution of practical tasks in work contexts. This change reflects the industry pressure for prototypes to demonstrate the ability to generate real and consistent productivity.
Digital Vision
Anderson Moreira, Professor of Control and Automation Engineering and coordinator of the robotics team at Mauá Institute of Technology, was interviewed by Olhar Digital about the World Humanoid Robot Games and the progress of robotics in China over the last decade.
According to Anderson, China has converted robotics competitions, which were previously limited to university events, into commercial showcases supported by government subsidies and extensive media coverage. He questioned how this boosts the sector and affects the global competitiveness of humanoid robotics.
Anderson explained that these competitions have stopped being just technological demonstrations and have started functioning as vast public engineering laboratories. By having dozens of manufacturers performing the same activities under standardized conditions, the limits of each solution become evident, such as autonomy, energy consumption, stability, precision, manipulation capability, and fault resilience. This generates an accelerated cycle of testing, problem identification, component redesign, and new testing.
In the 2026 World Humanoid Robot Games, for example, in addition to sports events, there will be industrial assembly challenges, material handling, cable connection, electric vehicle recharging, and fine manipulation, with over 40% of the events requiring fully autonomous operation.
However, China's greatest advantage lies in integrating these competitions with a vast industrial ecosystem. In 2024, China installed about 295,000 industrial robots, representing 54% of all new global installations. Additionally, for the first time, Chinese manufacturers were responsible for the majority of industrial robot sales within the country itself.
This scale fosters suppliers of motors, sensors, batteries, controllers, and power systems that can also supply humanoid manufacturers. Anderson Moreira emphasized that a competition demonstrates technological capability, but not industrial maturity.
There is also a clear industrial policy. China's Ministry of Industry and Information Technology has set a goal to put humanoids into mass production and create an internationally competitive ecosystem by 2027. In 2025, Reuters reported over US$20 billion allocated to the sector through various government programs, including regional funds, incentives, and public procurement. The same analysis indicated that the Chinese supply chain can already provide up to 90% of the components needed for a humanoid, and that only in 2024 did 31 Chinese companies present 36 distinct models.
This pressures global competitiveness by simultaneously reducing development time, component cost, and the gap between prototype and production. While the United States, Europe, Japan, and Korea may maintain leadership in certain technologies, they are now competing against a Chinese ecosystem capable of testing multiple architectures in parallel and rapidly industrializing the most effective ones.
However, it is important to note that a competition only demonstrates technological capability, not industrial maturity. A robot completing a test once is drastically different from executing the same task thousands of times, over months, with safety, low failure rates, and lower cost than human alternatives or traditional automation. The true future battle in humanoid robotics will be converting demonstration performance into reliable productivity.
Technical Challenges for Practical Applications
Although robots have made great strides in mobility in running tests, robotic hand engineering for fine gestures and precision constitutes one of the biggest obstacles in the industry. Anderson Moreira detailed the main technical barriers preventing these machines from leaving demonstration stages and operating efficiently in factories or homes.
The hand is considered one of the most complex systems to replicate in a humanoid, as it concentrates highly sophisticated mechanics, actuation, sensing, and control in a reduced space. Depending on the biomechanical model, the human hand has about 20 or more degrees of freedom, many of which must operate concomitantly.
In advanced robotic hands, this implies fitting multiple motors, transmissions, or tendons, sensors, and electronics into dimensions close to a human hand, while maintaining strength, speed, precision, low weight, and mechanical resistance. Recent research confirms this dilemma: increasing joints and actuators increases dexterity but also intensifies mechanical and control complexity.
The second major challenge is tactile perception. To handle a fragile object, connect a cable, or pick up a bottle, vision alone is not enough. The robot needs to know where contact occurred, in what direction the force is being applied, and whether the object is slipping. Humans make these adjustments instinctively; in robots, this requires tactile sensors distributed across the hand and algorithms that translate large volumes of data into real-time digital commands.
A study published in Nature Machine Intelligence, for example, covered 70% of the palm surface of an experimental hand with high-resolution sensing and demonstrated significant improvement in robustness when control utilized tactile information. The fact that this is still a scientific highlight illustrates how far we are from achieving this capability in an accessible, robust, and industrialized manner.
There is a third hurdle: contact control. Walking involves challenging but predictable dynamics. The hand, however, faces almost infinite scenarios: the object can be rigid or malleable, dry or slippery, slightly misaligned, or change orientation during manipulation. Applying too much force can break the item; applying too little can cause it to fall. This demands the continuous integration of vision, touch, finger position, and force control. Recent studies on anthropomorphic hands indicate that systems based solely on vision struggle with intense contact tasks because they cannot adequately regulate these forces.
The fourth bottleneck lies in the robot's intelligence itself. Artificial Intelligence models can learn specific tasks with thousands of examples, but it is difficult to ensure that the skill persists when the object, lighting, friction, or position changes. Unlike Large Language Models (LLMs), which are trained with vast amounts of text and images from the internet, the data for training robots must come from physical interactions, requiring a system to execute, record, and repeat real tasks. For this reason, China has created dedicated data collection centers where hundreds of operators train robots for long daily hours.
Finally, there are less conspicuous but commercially decisive requirements: durability, energy autonomy, ease of maintenance, safety, and cost. In a factory, it is not enough for the robot to connect a cable in a demonstration; it must repeat this throughout a full shift, detect failures, and recover autonomously. To work near people, there are rigorous safety and certification standards, such as the review of the international standard ISO 10218 in 2025, which deals with the safety requirements of industrial robots in production systems.
For this reason, it is believed that adoption will first grow in more structured industrial tasks—such as logistics, inspection, line replenishment, and material handling—and only later advance to very fine manipulations and domestic environments. The pending leap is not just making a robot perform a difficult task, but making it perform thousands of diverse tasks, with varied objects, safely and repeatably.
}, {