Robots dance together in the dark kitchen. How do humanoid robots move from science fiction to reality?

Figure robot (Source: Figure)

In the predictions of science fiction movies, the symbiosis between humans and robots always begins with a silent revolution - those once clumsy mechanical arms begin to think like humans, and the cold metal joints learn to dance with the scenes of life. In February 2025, this scene seems to be quietly playing out in reality.

The new generation of humanoid robots launched by Figure, an American startup company, is equipped with a vision-language-action (VLA) model called Helix, which greatly enhances the robot's perception, control, and understanding capabilities, allowing two robots to pass items tacitly without rehearsal . With its almost sci-fi intelligent collaboration capabilities, it brings the mechanical butlers in home scenes from fantasy into reality.

Humanoid Robots: The Ultimate Technology

Stills from the movie Metropolis

The concept of robot first appeared in the play "R.U.R." written by the famous Czech writer Karel Čapek in 1920. Maria, the android in the 1927 film "Metropolis", carried humanity in a mechanical body, becoming an early prototype of cyberpunk aesthetics.

A humanoid robot is a robot with human-like appearance and behavior that is made using artificial intelligence and robotics technology. It integrates cutting-edge achievements in multiple disciplines such as mechanical engineering, electronic technology, computer science, and artificial intelligence.

NASA's Valkyrie (R5) robot

Image source: NASA

The core technology of humanoid robots consists of four parts:

Perception system: refers to the ability of humanoid robots to perceive the environment and obtain external information through sensors and other devices, such as visual perception, auditory perception, tactile perception, etc.

Intelligent decision-making: refers to the use of artificial intelligence technologies such as machine learning, deep learning, and neural networks to enable robots to make autonomous decisions and judgments based on situational and task requirements.

Human-computer interaction: refers to the functions that enable humanoid robots to better communicate and interact with humans, such as language recognition, emotion recognition, etc.

Motion control: refers to the posture control, gait planning and balance control of the robot through the use of various sensors and algorithms.

The history of the development of humanoid robots is an epic story of humans trying to replicate themselves with machinery. From the gear fantasy of the steam age to the neural network revolution of the AI ​​age, the breakthroughs in each technological node coincide with the predictions and revelations of science fiction works.

01

Early development stage (1970s-2000):

Early humanoid robot models were simple in appearance, with only rudimentary human torso and limb shapes. The functions they could perform were also very limited. They could only perform simple walking and hand movements, did not yet have interactive capabilities, and had a low level of intelligence.

WABOT-1 robot (Photo source: Waseda University)

For example, in 1973, Waseda University in Japan developed the world's first humanoid robot WABOT-1, which could only perform tasks such as carrying objects, and its intelligence level was equivalent to that of a one-and-a-half-year-old baby. At the same time, the autonomous walking robots P2 and ASIMO developed by Honda in Japan were only continuously optimized in terms of walking and balance.

02

Highly integrated development stage (2001-2011):

This stage is the beginning of the primary perception function of humanoid robots. Due to technological breakthroughs, humanoid robots in this period can achieve limited interaction with the external environment and have improved freedom of movement. Although the application scenarios are relatively simple, mainly used for exhibitions and entertainment, there are indeed great advances in motion control and human-computer interaction.

Honda Robotics Development History and the Third Generation ASIMO (Source: Honda)

For example, in 2003, Sony's QRIO robot was equipped with voice recognition and face recognition functions, and could recognize 10 facial expressions; in 2011, Japan's Honda launched the third-generation ASIMO, which has the ability to automatically judge and act by using sensors to avoid obstacles, and can also use five fingers to sign language or pour water from a kettle into a paper cup; ASIMO can not only go up and down stairs and serve tea, but also conducted the Detroit Symphony Orchestra in 2014.

03

High dynamic motion and interactive capability improvement stage

(2012-2020):

The research and development focus of this stage is to enhance the movement or interaction capabilities of humanoid robots. Reinforcement learning technology allows humanoid robots to continuously optimize their movements and behaviors in interaction with the environment, and emotion recognition and language interaction technologies are also significantly improved.

POPPY robot development history (Source: Inria)

For example, the bipedal humanoid robot Atlas, developed by Boston Dynamics in the United States in 2013, has super high athletic ability. It can skillfully complete vertical jumps, handstands, cross obstacles, backflips, and even dance with the Spot robot. It has parkour functions that involve both hands and feet.

In 2016, the first open-source humanoid robot POPPY developed by the French Inria Flower Laboratory was launched on the market. It has demonstrated excellent adaptability in many fields such as education, scientific research, culture and art. In the field of education, POPPY can be used as a teaching tool to help students understand the principles of robots and programming knowledge more intuitively; in cultural and artistic creation, artists use POPPY's unique expressiveness to create many novel works of art.

04

Highly intelligent development stage (2020 to present):

With the help of modern artificial intelligence, computer vision systems and other advanced technologies, humanoid robots are moving towards intelligent advancement. Today's humanoid robots have greatly enhanced perception and cognitive abilities, most of which have highly bionic torso configurations and anthropomorphic motion control, and have made significant progress in weight, flexibility and other aspects, greatly enhancing their practicality.

Ameca Robotics Development History (Source: Engineered Arts)

For example, after Engineered Arts in the UK connected Ameca to GPT-3/4, it was reborn. Ameca is equipped with advanced sensors and has facial and multiple voice recognition capabilities. It can interact with humans naturally, not only can it keenly detect human emotions, but also can convey information through rich expressions and gestures.

In some high-end technology experience places, Ameca serves as a receptionist, providing services to visitors with cordial communication and interaction, which leaves a deep impression. The mobility, dexterity, balance and real-time processing capabilities of Tesla's Optimus Gen2 humanoid robot have also been significantly improved due to the intervention of the AI ​​big model.

Unitree H1 robot (Source: Unitree Technology)

Although the development of humanoid robots in China started relatively late, with research only beginning in the 1990s, in recent years, the research and development of humanoid robots in China has made significant progress, with many companies and scientific research institutions actively participating in it and continuously exploring technological innovation and product applications. Yushu Technology is a leader in the field of humanoid robots in China.

On the Spring Festival Gala stage, 16 humanoid robots Unitree H1 from Unitree Technology performed "Yang BOT" with dancers, stunning the audience with the "cyber yangko". These robots have 19 joints and their arms have an additional 3 degrees of freedom, which enables difficult movements such as handkerchief rotation and throwing and recycling.

Behind this dance named "YangBOT" is the precise coordination of 19 joints, the AI ​​mapping of each frame of movement, and the breakthrough progress of bipedal robots in the field of complex dynamic balance, which fully demonstrates the profound strength of Yushu Technology in the research and development of humanoid robots.

The development prospects of humanoid robots are full of expectations. With the continuous innovation and breakthroughs in artificial intelligence, material science, sensor technology and other fields, humanoid robots will become more intelligent, flexible and powerful.

In terms of intelligence, they will have stronger learning and reasoning abilities, and will be able to understand and handle more complex tasks and situations. Through deep integration with technologies such as the Internet of Things and big data, humanoid robots can acquire and analyze massive amounts of data in real time, making more accurate decisions and more efficient actions.

In terms of athletic ability, new materials and mechanical designs will make humanoid robots' movements smoother and more natural, enabling them to adapt to various complex terrains and environments and complete more difficult movements.

Figure robot (Source: Figure)

As humanoid robot technology continues to mature, its application scenarios are becoming increasingly extensive . In the industrial field, they can undertake some dangerous, repetitive or high-precision work; in the medical field, humanoid robots can assist doctors in surgery, and with their high-precision operating capabilities, they can reduce surgical risks; in terms of home services, humanoid robots can serve as housework assistants, sweeping the floor, cleaning windows, washing dishes and other housework can be easily completed; in the field of education, humanoid robots, as innovative teaching tools, can impart knowledge in a vivid and interesting way.

It can be seen that humanoid robots have great potential in the future, but they still face many challenges.

1. High computing power: The posture, speed, and strength of humanoid robots all require high computing power, and the current computing power faces huge challenges.

2. Weak technology: The current basic technical capabilities are still very weak. The existing software algorithms are not sufficient to support the large-scale application of humanoid robots, and major technological innovation is still needed.

3. High cost: High R&D costs, high losses, and high R&D investment are the norm in the humanoid robot industry.

4. Insecurity: Data security capabilities and systems are insufficient, and face security and privacy issues.

Connor, the protagonist of the game Detroit: Become Human

From the simple movements of WABOT-1 to the free interaction of Ameca, humanoid robots are entering our daily lives as a "technological miracle". They are the evolution of HAL 9000 in "2001: A Space Odyssey" and the prototype of Connor in "Detroit: Become Human". However, the ultimate goal of technology is not to replace humans, but to become a collaborative partner in exploring the unknown, like TARS in "Interstellar".

In the future, when humanoid robots are truly integrated into society, humans may redefine "humanity" - not based on biological characteristics, but on creativity, empathy and ethical choices.

Perhaps, just like all the classic science fiction fables, humanoid robots will eventually become a mirror that reflects the essence of human beings: when they learn to adjust the strength of their fingertips to prevent bumps when handing out an apple, can we treat our fellows with the same gentleness? When they break through the Turing test, how should the boundaries of "consciousness" that humans are so proud of be redefined?

The answer may be hidden in the laboratories of the next decade, or it may have already been written in Asimov's laws of robotics and the rainy night monologue in "Do Androids Dream of Electric Sheep?" The only certainty is that the final chapter of this mechanical awakening is destined to be written by human civilization and machine intelligence.

References:

https://www.ccidgroup.com/info/1207/41123.htm

https://www.hanspub.org/journal/paperinformation?paperid=86460

https://www.figure.ai/news/helix

https://en.wikipedia.org/wiki/Atlas_(robot)

https://builtin.com/robotics/humanoid-robots

https://en.wikipedia.org/wiki/Humanoid_robotAuthor: Yang Yuxin

Planning: Liu Ying, Zhang Chao, Li Peiyuan, Yang Liu

Reviewer: Fu Changyi, Associate Professor, Nanjing University of Technology

Chairman of the Science Fiction Committee of Jiangsu Science Writers Association