Over the past year, robots have become more visible to the naked eye.
Bipedal humanoid robot, wheeled robot, four-legged robot dog…… There are more and more companies that make robots. Foreign countries have “to mass produce Optimus Prime robot in 2025” Tesla, last year just got OpenAI investment humanoid robot startup 1X, “AI godmother” Li Feifei’s first venture is also a robot-related space intelligence. In addition to the robot LABS owned by Tencent and Millet in China, Zhiyuan, Yushu, Horizon, Youbi and more startups are joining the entrepreneurial boom of robots.
According to Tianeye data, interface news incomplete statistics, in 2023 a year, there were 23 financing cases in the global humanoid robot field, with a total amount of 5.47 billion yuan, the highest level in a decade. In 2024, the investment boom continued, and in the first half of the year, only domestic humanoid robots invested and financed more than 10 cases, and the total financing exceeded 2 billion yuan.
By the end of 2024, there are more than 80 domestic humanoid robot machine companies, while the number is close to 150 globally. In the next year, with the increasing number of robot companies, humanoid machines will usher in a “hundred machine war”, and in the increasingly crowded robot circuit, where the next development of the industry will go.
The concept of embodied intelligence is hot, but the inflection point has not yet arrived
In the past two years, an important direction of investment in humanoid robots with fire is embodied intelligence.
Whether in science fiction or TV dramas, the high-tech intelligent evolution direction represented by robots has left a deep image for ordinary people. Since the ChatGPT large language model triggered the generative AI revolution, what the next form of AI will look like has become a hot topic in all walks of life. Robots do the torso, AI does the brain, and “embodied intelligence” has naturally become a new story in the robot industry.
Unlike software platforms like ChatGPT, embodied intelligence emphasizes robots as hardware vehicles for AI, both of which develop and demonstrate intelligence through direct interaction with the physical environment. With the addition of the “body,” intelligence is not just the result of information processing, but also the ability to perceive, act, and adapt to the real physical world.
Embodied intelligence also became one of the hottest topics in the AI industry last year. At the Nvidia GTC Conference held in March, Nvidia released a number of products such as models, chips, and software platforms dedicated to the development of humanoid robots. CEO Jen-Hsun Huang took the stage with more than 20 robot products from companies such as 1X, Agility Robotics, Boston Dynamics, Figure AI, and Uki, and predicted in his speech that “the next wave of artificial intelligence will be embodied intelligence.”
Turing Award winner Yao Qizhi uses “brain-cerebellar-body” as an analogy to embodied intelligence. In his view, humans themselves are an ideal embodied agent. Current AI systems, such as ChatGPT, are primarily focused on language processing capabilities. As the next form of AI, embodied intelligence first needs hardware as a “body”; Sensors and other actuator devices as the “cerebellum”, responsible for processing perceptual information, control the body to complete complex tasks, AI as the “brain”, responsible for logical reasoning, decision-making and planning, so that the intelligent body can use natural language to communicate with other agents and the environment.
For the robotics industry, creating a universal humanoid robot in any scenario is the ultimate goal of embodied intelligence, but in terms of the technical level of the industry, there are still many practical difficulties to achieve this goal.
Two people in the robotics industry told reporters that the realization of embodied intelligence relies on the integration of multiple technical fields, including language cognition, behavioral intelligence and visual intelligence, and the process complexity is extremely high, and the existing level is far from the threshold requirements of embodied intelligence.
Take the data conundrum that has plagued the robotics industry for years. High-quality data is the foundation of machine autonomous learning, but data in the real world is not directly available for robot development, and training robots to complete different tasks requires a corresponding amount of data. Missing data can directly affect a robot’s performance: According to a study by Nvidia, a robot trained on a small amount of data has a very low success rate on a specific task, which is only 0.7% according to the team’s statistics, but after training with a rich data set, the success rate can jump to 76%. Previously, OpenAI disbanded its robotics team due to a data shortage. After the advent of ChatGPT, the robotics industry is also exploring the use of model generation to solve the problem of missing data, and there is no fundamental breakthrough.
The mainstream view of the industry believes that if the advent of ChatGPT is the “iPhone moment” of the AI industry, the inflection point of humanoid robots on the road to embodied intelligence has not yet arrived, and the industry is still on the eve of the outbreak.
Manufacturers promote price cuts, and “a robot” is still far away
The robot industry has experienced many booms and bubbles before, and high costs have been one of the important reasons why robots have not been able to land on a large scale. The price of early humanoid robots is generally higher, Honda’s humanoid robot ASIMO, Boston Dynamics’ tlas, the industry conservatively estimated the cost of more than $2 million.
After more companies poured into the robot industry, price cuts became a major trend in recent years.
In 2021, when Tesla publicly displayed its Optimus Prime humanoid robot for the first time, Musk gave an ambitious mass production plan – start mass production in 2025, achieve mass production in 2026, the final price of Optimus Prime may be as low as $20,000, and the future hopes to produce more than one million units – in the future blueprint described by Musk, “At some point in the future, robots will outnumber humans in a ratio of more than 1:1.”
In the past two years, robot manufacturers have been looking for ways to control costs and promote price cuts. In the total cost composition of a humanoid robot, hardware accounts for about 70%, software accounts for 30%, and the hardware motor and reducer sensor are major cost items. According to Macquarie, a market research firm, the total cost of hardware for an early humanoid robot, including key components such as sensors, servo motors and precision reducers, is about $40,000.
In May last year, domestic robot company Yushu Technology launched the second generation of humanoid robot G1, with a starting price of only 99,000 yuan, compared with its previous generation of humanoid robot H1’s $90,000 price, about 80% lower, causing a stir in the industry. According to Wang Xingxing, founder of Yushu Technology, the company reduces production costs by independently developing core components and optimizing product design.
After Yushu, another startup, Zhongqing Robotics, launched a humanoid robot, PM01, priced at just 88,000 yuan, continuing to help drive down prices.
It is believed that due to the increase in the number of companies pouring into the robot industry, market competition has intensified, and this competitive pressure will push the price of humanoid robots down. Since robot companies in the market almost all rely on financing to survive, taking this price reduction strategy to seize the market, in turn, can win more financing support for the company.
According to the New York asset management company ARK Investment (ARK) research report predicts that in order for humanoid robots to be able to achieve mass adoption, their cost will need to fall by another 50% to less than $20,000, about half the current level of the industry as a whole.
But robots have yet to agree on a price war. Last year, the humanoid robot company Zhiyuan Robot, which completed several rounds of large-scale financing, clearly stated that it does not participate in the price war. The company’s star founder “Zhihui Jun” Peng Zhihui publicly stressed that the goal of Zhiyuan robot is not only to develop advanced technology, but also to achieve the popularization of intelligent robots through open source and open cooperation, rather than only to obtain market share by lowering prices in the short term.
Uki, which has led the charge, is also wary of slashing prices on a large scale. In an interview with the interface news in August last year, Wang Xingxing, founder of Yushu Technology, also stressed that the robot industry is far from reaching the commercialization stage, and the total number of robots on the market is still very small. The annual shipment of the head company is not more than 100 units, and most companies only have dozens of units, which is not enough to support the price war. “If we set prices too low, it’s bad for the industry and bad for the company.”
Now it seems that the next “hundred machine war” is difficult to directly cause large-scale price cuts. The robot industry is more willing to attribute the entrepreneurial boom of the past two years partly to Musk’s topic, and partly to the explosion of ChatGPT at the end of 2022, which is still far away from the commercialization stage of “a robot with a human hand”, the technical route still has many vague areas, and the cost price still has room for further exploration. But the future of the industry is optimistic, with Wang Xingxing using an analogy to describe the current situation: “Last year is like two years before the birth of ChatGPT.”
