Evolve! Humanoid Robot
2024-08-12
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
Evolve! Humanoid Robot
Banyuetan reporter Zhang Manzi
Recently, the humanoid robot track is crowded with giants. This may not be the most futuristic track, but it is the most intuitive. my country's humanoid robot "first stock" was born just half a year ago, and many start-ups have been working non-stop to obtain financing and enter the "unicorn" list. With the support of artificial intelligence big models, the new members of the robot show new trends of interactivity, autonomy, and versatility.
"The next wave of AI is physical AI. Robots will be integrated into our daily lives, and everything will be robotized." In early June this year, the founder of Nvidia made a prediction. Imagine, on a hot summer day, after a day of work, we tell the robot "I will be home in half an hour", and the robot immediately turns on the TV and air conditioning and prepares a table of delicious food. How comfortable life will be.
Are humanoid robots necessarily the ultimate form of general artificial intelligence? Not necessarily.
Where do the flowers on the high mountains bloom?
There may be no mobile phones in the future, but there cannot be no robots. In other words, compared to toys, robots are definitely “tools”.
As toys, robots are good at such fancy things as chatting, singing, dancing, and playing Go; while humanoid robots as tools are now entering practical and urgent scenarios such as serving tea, frying eggs and rice, carrying heavy objects, and working at high altitudes.
Not long ago, UBTECH's industrial version of the humanoid robot WalkerS entered NIO's final assembly workshop for "practical training" and became a trainee factory worker with a keen eye. Its dexterous hands accurately affixed the car logo to the front of the car, its agile feet led its torso to coordinate and automatically move to each seat to perform seat belt tension testing, and its smart eyes focused on the doors and body surface for defect detection.
Industry will be the first mainstream scenario for large-scale application of humanoid robots. Zeng Xuezhong, senior vice president of Xiaomi Group, predicts that future smart factories will be composed of three types of labor: 70% automated equipment, 20% humanoid robots, and 10% humans. Humans are responsible for value judgment and mining, while humanoid robots are responsible for flexible manufacturing and cross-system collaboration. The collaborative work of the three types of labor will redefine "workers" and "factories."
In the smart factories of the future, humanoid robots with environmental complexity as the vertical axis and operational complexity as the horizontal axis will diverge from the origin to the first quadrant: from the fixed environment of a small space to the complex environment of a large space; from automated direct operation to the use of structured work for assembly operations, and then to the flexible use of multiple tools for general operations.
On May 24, at the on-site experience area of the 7th Digital China Summit, the super-bionic humanoid robot "Xiaoqi" greeted the audience. Photo by Wei Peiquan
At present, humanoid robot companies such as Zhiyuan Robot and Xiaomi Robot have begun to develop industrial humanoid robots that can realize functions such as automobile chassis assembly, appearance inspection, aging testing, gear column oiling, and material handling.
In addition to industry, agriculture is also a scenario with both demand and implementation potential. Take Shandong, a major agricultural province, for example. Every apple picking season requires a lot of manpower. The apple trees on the Shandong Peninsula have tall crowns, high temperature and high humidity, and high risks in picking. In fact, the amount of labor required in seedling cultivation, transplanting, management, transportation and other links is quite huge.
Farmland operation robots, fruit and vegetable picking robots, and facility vegetable production robots will outline a future of smart agriculture for major agricultural countries.
At present, the Netherlands has developed a rice seedling production robot, Italy has developed a leafy vegetable harvesting robot, Spain and Japan have strawberry picking robots and tomato picking robots respectively. However, due to the terrain restrictions such as mountains and hills, the complex environment of weeds and special vegetation, and the disordered arrangement of fruits, the difficulty of visual recognition has increased, which has prolonged the landing time of humanoid robots.
How to achieve embodied intelligence?
Robots are not new, but what is so great about this generation of robots?
The underlying hardware and motion control are the original technologies of traditional robots, while the embedded artificial intelligence model is the product of the new era and the most imaginative part of this generation of humanoid robots. The artificial intelligence model injects a "soul" into the robot, giving the machine a "brain" that can learn, make decisions, and reason autonomously. This is the most significant difference between humanoid robots and traditional robotic arms, and it also allows people to see the possibility of robots becoming universal.
In the past, traditional dedicated robots or robotic arms that only performed a single operation, such as welding robots and grinding robots, were essentially automated equipment that only executed fixed programs written in advance by humans.
After being equipped with a large model, the humanoid robot has three levels of human-like abilities: the most superficial limb movement ability, the intermediate multimodal perception ability, and the higher-level decision-making and control ability. These three abilities come from the three major components of the humanoid robot: the execution system, the sensor system, and the control system, which correspond to the three types of "organs" of the humanoid robot - the body, the sensory organs, and the brain and brain.
The key to humanoid robots is not the "shape" but the "brain" that directs the movement of the body. The "playing style" of robots has changed. In the past, robot entrepreneurs may have majored in automation or mechanics, but now humanoid robots carry the high hopes of moving towards embodied intelligence (referring to an intelligent body with a physical body that supports perception, movement, and interaction). Entrepreneurs who engage in humanoid robots must not only understand hardware, but also have accumulation in algorithms, vision, AI and other fields.
The revolutionary change that the iteration of the big model brings to robots is that it is not only applied to the understanding, reasoning, and planning of the robot's "brain", but will also be integrated into the perception and execution process involving the "cerebellum" and "trunk". With more powerful generalization capabilities, humanoid robots will be better able to interact with the physical world, further reducing deployment costs and application barriers. This is also the direction of embodied intelligence.
When will the evolution of machines enter our homes?
In recent weeks, predictions from across the ocean have emerged one after another: "In the future, all moving objects will be able to operate autonomously", "Humanoid robots will be as common as cars, and one person may have two humanoid robots, or even more"...
Although the global humanoid robot track is currently facing technical barriers, data bottlenecks, and mass production challenges, the aging and declining birthrate society is calling for humanoid robots more and more urgently. With the help of multimodal large models and the continuous evolution of AI agents, the growth path of humanoid robots will also continue to evolve from primary execution operations to intermediate knowledge learning and then to high-level wisdom formation.
The implementation of humanoid robots will start with entering one scenario (such as industry) and realizing multiple functions, then expand to integrating into multiple scenarios (logistics, agriculture, consumption, etc.), realizing complex functions, and then gradually entering thousands of households. The humanoid robots entering the home will be divided into two steps: the first step is to solve the need for companionship; the second step is to enter the service scenario. For example, when watching a game with you, they can understand whether you support Argentina or Portugal by observing your words and expressions, and then cheer with you.
Visitors touch a humanoid robot at the 2024 World Intelligent Industry Expo held in Tianjin on June 22. Photo by Li Ran
If Generation Z is the natives of the digital Internet, and the people born in the 10s and 20s, known as the Alpha Generation, are the natives of artificial intelligence and large models, then the next generation may become the natives of humanoid robots and their life partners.
At the Computer History Museum in Silicon Valley, just a few blocks from Stanford University, you can see the changes in computing and the iteration of hardware over the past century. The threshold for using hardware is getting lower and lower, and the market size is getting bigger and bigger. The trend of technology drifts back and forth, and the seats of technology giants are constantly changing. In the future, there will be a battle for the chairs between the giants of artificial intelligence software and hardware.
With the influx of new entrepreneurs in the field of humanoid robots, a huge market with an industry scale of more than 30 trillion US dollars by 2050 is unfolding before people. In big waters, there must be big fish. Practitioners hope that great companies can be born in China in this field, rewriting the history that 90% of high-end applications of robots in key industries are monopolized by foreign countries.
Looking back at the evolution of computing hardware from complex to simple, after humanoid robots, perhaps there will be simpler, more portable, and more miniaturized hardware to replace these robots that are similar in height, weight, and appearance to humans. Perhaps it will be a black box, a screen, or just a thin film. However, the direction of machine evolution will not change: machines will be more proactive in understanding humans, rather than teaching humans how to use and adapt to machines.
