I can't hold it in anymore. Is a humanoid robot just a machine plus a "half - human"? Does it have a physical form but no intelligence and rely entirely on "human intervention"?

The past year can be regarded as an acceleration year for embodied intelligence. The robot sports meeting last week fully demonstrated the ability of robots to dash forward - well, let's say run at full speed.

Even more astonishing, an engineer was pushing a wheelchair with a robot sitting on it.

These memorable scenes at the sports meeting are not only funny but also clearly reveal a fact: behind every successfully operating robot, there is always a human being.

Injecting "Soul" into Robots

The most typical and necessary scenario of "human - robot collaboration" is autonomous driving. Traffic safety is of utmost importance. Therefore, in the actual implementation process, many seemingly intelligent and autonomous robot devices still require manual remote control or intervention.

For example, Cruise, a subsidiary of General Motors, once disclosed that its driverless taxis need to trigger remote human assistance approximately every 6 to 8 kilometers. During fully driverless operations, about one remote assistant is on standby for every 15 - 20 vehicles.

However, General Motors announced the lay - off of the entire Cruise business group at the end of last year. In the United States, Waymo almost dominates the market. As early as 2020, Waymo deployed a team of human monitors on the outskirts of Phoenix, Arizona.

They assist the driverless taxi service through remote operation. Once the vehicle encounters an abnormal situation, the remote operator can send instructions to make the vehicle drive as required.

Similarly, companies like Luobo Kuaipao have also arranged safety officers so that when the autonomous driving system is unable to handle a situation, the vehicle can be taken over by a remote driver.

More broadly, in the fields of service robots such as warehousing and logistics, security inspection, and food and delivery services, there are also a large number of phenomena of remote manual operation. Many delivery robots will hand over control to real - life operators sitting in the back - end when encountering complex road conditions. Some commercial cleaning robots and security robots are equipped with remote monitoring centers, where staff can take over or guide the machines to complete tasks at any time.

Current robot technology is still insufficient to cope with the ever - changing real environment. Human experience and judgment are injected into robots through the network, becoming a timely supplement.

The "Human Loop" in Embodied Intelligence

Jensen Huang, the CEO of NVIDIA, has envisioned the future of "physical AI" many times, predicting that humanoid robots will be widely used in real - world applications - although at present, they still cannot do without humans.

According to a report by The Information, the small robot standing beside Jensen Huang in March this year also had a commander behind it. NVIDIA's advanced simulation training has enabled robots to have basic movement skills, but in real - world scenarios, it is still up to the people behind the scenes to decide when and where the robot should go to complete specific tasks.

Behind the intelligent evolution of many robot projects served by NVIDIA, there are countless times of repeated remote manual operations.

Although this "shadow labor" model is costly, it is considered an effective way to teach robots skills at present. For example, OpenWBT_Isaac realizes the full - body remote control of humanoid robots in virtual and real environments based on the Nvidia Isaac simulation platform, allowing operators to command robots to complete actions like playing a game.

In this process, NVIDIA plays the role of a "bridge - builder" - using powerful computing power and tools to continuously inject human experience into robot intelligence and gradually let AI take over more control. In Jensen Huang's vision, the ultimate stage of development may be "AI assisting humans in the loop", but at least for now, humans still firmly control the decision - making loop of robots.

Wang Xingxing, the founder of Unitree Robotics, also admitted that at present, to achieve the best performance, robots， often still need manual participation in scheduling.

The humanoid robots of Unitree Robotics performed a yangge dance collectively at the 2025 Spring Festival Gala on CCTV, causing a sensation. Of course, the complex dance steps and neat formations of the robots were mainly achieved through pre - arrangement and real - time remote control by engineers. After the performance, when the robots were going down the steps, someone needed to help them.

After the World Humanoid Robot Sports Meeting last week, Wang Xingxing revealed that although the H1 theoretically has the ability to run autonomously without a remote control, in order to pursue speed, they finally adopted a manual remote - control strategy in the competition. In other words, for the sake of performance in the arena, Unitree chose a more reliable human - control scheme rather than letting the robots play freely.

This reflects the current reality of technology: Robots can run on their own, but to run fast and steadily, humans are still more reliable commanders.

Although Unitree Robotics is at the forefront of robot commercialization, its product implementation still currently adopts a "human - machine hybrid" model: the machine is responsible for performing specific actions, while humans are responsible for planning routes and dealing with emergencies. For now, behind almost every running Unitree robot, there is a person holding a controller tightly.

Invisible Shadow Workers

Is it just for show when we talk about automation and robots, but actually still need human labor behind the scenes?

For a long time, human - robot collaboration has penetrated into many mature industries. There are high - precision fully automated procedures for airplanes and surgeries, but pilots and doctors still need to be present - human supervision can enhance the reliability of automation.

This results in a relatively high requirement for the ratio of personnel to machines to ensure safety. For autonomous driving, the Ministry of Transport requires that the ratio of remote safety officers to vehicles should not be lower than 1:3, that is, one safety officer for every three vehicles.

At the same time, to ensure the safety of driverless vehicles, remote supervisors may always need to be on high alert. A safety officer from Luobo Kuaipao once said in an interview that they need to work 8 hours a day, with a 10 - minute break per hour. An alarm will go off if a safety officer closes their eyes for more than three seconds.

If each robot still requires half or even more human labor to support it, the claimed significant efficiency improvement advantage of automation will be offset. When the scale of robots expands from dozens or hundreds to thousands or even tens of thousands, the remote operation team itself will also expand into a large support center.

On the one hand, there is the question of whether real cost - reduction and efficiency - improvement can be achieved. On the other hand, when drivers become safety officers, there are changes in salary and working conditions. Are these changes reasonable and cost - effective for workers?

In a sense, behind every seemingly smart robot, there stands an invisible shadow worker. When technology advances rapidly and AI becomes more and more powerful, perhaps one day these unsung heroes will retire.

But before that, they will still fight side by side with robots for a long time. The era of robots is not a pure machine era but an era of symbiotic collaboration between humans and machines.

This article is from the WeChat official account "APPSO". The author is APPSO, which discovers tomorrow's products. It is published by 36Kr with authorization.