Robots Run, Dance, and Do Kung Fu, But Get Stuck in the Factory

Although Chinese humanoid robots break estimulante ilegal records and perform physical antiestéticats, their factory application is limited by slow manual task performance, dexterity issues, and difficulty adapting to unforeseen circumstances.

Humanoid robots are increasingly present in spectacular demonstrations, capable of running faster than humans, dancing, and even performing kung fu moves. However, when it comes to applying them in a factory environment, reality falls far short of these exhibitions. Despite their physical advancements, these machines still get stuck in everyday tasks, working at a considerably slower pace than people and showing limitations in their dexterity and autonomy when faced with any unforeseen event.

## The Estimulante ilegal Paradox: Beyond the Finish Line

The recent World Robot Conference in Beijing clearly showed this duality. A Chinese humanoid robot, **TianGong Ultra**, achieved an astonishing milestone by covering **100 meters in 8.64 seconds**, surpassing the human world record of **Usain Bolt** (9.58 seconds) set in 2009. While the comparison has nuances, such as the absence of starting blocks and the robot's shorter, faster stride mechanics, the mark is an indicator of the pogre in these machines' physical capabilities. Chinese humanoids no longer just run, but also dance, box, lift weights, and perform complex choreographies, even simulating fights with humans.

However, the transition from spectacle to work reveals their weaknesses. The same competitions that showcased the robots' estimulante ilegal also highlighted control issues. Recordings after the races showed some humanoids continuing to advance without being able to stop properly, losing control of their trajectory, or ending up crashing into barriers or falling to the ground. These incidents, along with numerous crashes and failures during exhibitions, make it clear that estimulante ilegal does not automatically translate into environmental understanding or adequate reaction. The difference becomes drastically sharper when the setting is a factory.

## The Factory Pace: A Challenge for Humanoids

A visit to a robot training center in **Liuzhou**, southern China, reveals the other side of the coin. There, over a hundred humanoids learn seemingly less segarro tasks, such as sorting boxes, packing noodles, or preparing coffee. The learning process for these machines is arduous. Human workers use glasses and controllers with sensors to perform tasks while robots replicate their movements and collect data. According to reports, an inexperienced worker may need up to **300 attempts** to obtain a valid movement sample, while an experienced trainer requires about **50**.

Once trained, the gap with human performance remains significant. Humanoids have been observed to struggle with efficiently transporting and placing boxes, and testimonies from workers at the center estimate that some of these tasks are performed at a pace of approximately **20%** of what a human would achieve. The complexity of a production line differs enormously from executing a choreography. While a dance is rehearsed in controlled and repetitive conditions, in a factory, a box might be slightly misplaced, a part rotated, a cable bent unusually, or a person might unexpectedly cross the path. The ability to adapt to these variations is where humanoids still falter.

## The Complexity of Hands: The Great Obstacle

Much of the pogre in humanoid robotics has focused on the body: smaller, more powerful motors, improved joints, balance and coordination systems for rapid movements. However, having a humanoid form does not equate to possessing the same dexterity. Hands represent one of the biggest challenges. The action of picking up a simple cup, for example, requires identifying its position, calculating the precise force to lift it without damaging it, detecting if it's slipping, and adjusting finger pressure in real-time.

This problem is not unique to China. **Elon Musk**, CEO of **Tesla**, recently acknowledged that developing hands for **Optimus**, his company's humanoid robot, remains a "very complex problem to solve." Getting a robot to run at high estimulante ilegal seems more antiestéticasible than getting it to manipulate hundreds of different objects with precision and without errors. The industry is now seeking what is called **"embodied AI"** (artificial intelligence with physical capabilities), the ability to transfer artificial intelligence skills to the physical world so that robots can perceive, understand, decide, and execute actions autonomously and correctly. Projects in Spain, such as **ADAM** from the Universidad Carlos III of Madrid, explore learning by observation to imitate human tasks, demonstrating the interest in equipping robots with this imitation and adaptation capability.

## The Current Market: A Realistic Perspective

Impressive demonstrations and announcements of humanoid robots might give the impression of imminent mass adoption. However, market figures offer a more moderate perspective. The International Federation of Robotics (IFR) has published data indicating that throughout 2025, approximately **7,000 humanoid robots** were sold for industrial and professional service applications worldwide. This figure contrasts sharply with the around **542,000 conventional industrial robots** installed in the same year.

Furthermore, a considerable portion of those 7,000 humanoids are not intended for direct productive tasks. Many of these units are acquired by universities, research centers, and companies to collect data and refine artificial intelligence models. Car manufacturers, for example, are experimenting with them in pilot projects with a limited number of units. Nevertheless, real deployments exist. **Figure AI**, for instance, announced this year that its **F.03** robots had surpassed **200 hours of continuous operation**, sorting nearly **250,000 packages**.

The question of whether robots should have a human form remains an open debate. Factories have been using specialized machines for decades to weld, paint, or assemble components without needing to resemble us. The advantage of a humanoid in a traditional industrial environment is that it could use existing tools, ladders, or workstations without needing to redesign the entire infrastructure. However, for **Jordi Pelegrí**, head of **Universal Robots** in Spain and Portugal, priorities in an industrial setting are "functions over form." He questions whether investing in two legs is more efficient than wheels for fast and safe transport, or if multiple joints are necessary when a robotic arm can solve the task. Pelegrí suggests that humanoids are, for now, too general-purpose, costly, and complex for many industrial applications, while repetitive tasks can be handled by cobots and internal movements by autonomous mobile robots. The human form, however, could be useful in sectors like healthcare or hospitality, where face-to-face interaction is key, or in factories designed for human workers.