Connecting silicon-based brains represented by large models with joints equipped with various sensors, enabling robots that "write a piece of code to do something" to have the ability to autonomously "integrate knowledge and action", this is the most popular humanoid robot since last year. International market research firm Strategies MRC predicts that from 2021 to 2028, the size of the humanoid robot market will increase from $1.5 billion to $26.4 billion, a growth of nearly 18 times. Seemingly within reach, when will it become a daily consumer product like a mobile phone? At the 5th China International Conference on Innovation and Development of Large Models and Intelligent Robots, experts believe that the entry of humanoid robots into households still needs to overcome three obstacles: joints, datasets, and application scenarios. The entry may start with the leasing model. Why is a humanoid robot known as the most difficult robot in the industry due to its expensive joints, with a drive costing 20000 yuan per unit? Therefore, like humans, it is composed of the brain, cerebellum, skin, five fingers, joints throughout the body, and delicate torso, each with a large amount of technical details. "The brain corresponds to the intelligent system, the cerebellum corresponds to the control system, and the facial features correspond to the perception of the robot, while execution is carried out through joints," said Zhou Meijie, Vice President of Shanghai Robotics Industry Technology Research Institute. At present, the highest cost for humanoid robots is the joints, which are mainly composed of motors, drivers, reducers, encoders, force sensors, etc., accounting for about two-thirds of the overall cost. For example, servo motors used to ensure the flexibility of robots have an average selling price of around 2000-3000 yuan in foreign countries. On average, a humanoid robot requires 20-30 units, and the cost of optical motors can reach 100000 yuan. In terms of hardware, the core issue faced by humanoid robot companies is the driver. Zhou Meijie revealed that the driver current required for full-size humanoid robots is around 300 amperes, but currently mainstream drivers can only achieve around 120 amperes. At present, most of the drivers for humanoid robots come from an Israeli manufacturer that can provide 200 amperes of current, and such a driver costs 20000 yuan. Aiming at this pain point, Shanghai Robotics Industry Technology Research Institute is developing high-power drivers with an expected current of up to 300 amperes, and the cost can be reduced to around a thousand yuan. Irregular actions and lack of large-scale standardized datasets. With the rise of artificial intelligence large models, specialized large models will be deployed within humanoid machines to understand human commands and complete various tasks. The raw material for training large models of robots is datasets. "The current lack of large-scale standardized datasets in China is a common problem." Zhou Meijie said. The so-called standardized dataset includes a series of behaviors such as data collection, data uploading, automated annotation, model training, large-scale regression testing, application validation, and data augmentation, which require a deep understanding of the industry and robots. In April of this year, led by the Shanghai Institute of Robotics Industry Technology and in collaboration with research teams from Shanghai Jiao Tong University, Fudan University, Tongji University, as well as enterprises such as Fourier Intelligence and Zhiyuan Robotics, a humanoid robot dataset construction project was launched. The dataset will be open to domestic enterprises and research teams to train large models of humanoid robots, enabling robots to have developed brains and cerebellum Zhou Meijie said. The embodied intelligence enterprise Feixi Robotics, which focuses on robotic arms, is also engaged in related research and defines it as "original capability". Hu Xiaoping, Vice President of Feixi Technology, explained that when analyzing the underlying functions of the robotic arm, Feixi extensively referred to human behavioral research, breaking down complex human actions into the most basic 64 primitive abilities. Subsequently, various complex operations in the interaction with the entire physical space can be achieved through combination. At present, the deviation of the force distance sensor of the non Xi robotic arm can be controlled within three thousandths, whether it is cutting chicken chunks, automatic plate placement, or removing vegetable roots for Fuling pickled vegetables, it can be easily handled. "The lower the price, the wider the market application, and the faster humanoid robots can enter thousands of households," said Zhang Yongwen, General Manager of Shanghai Feiyun Robot Technology Co., Ltd. At the just concluded China (Shanghai) International Technology Import and Export Fair, the appearance of the Feiyun robot holding a small fist and taking small steps left a deep impression on the audience. The reason why we focus so much on the application side is that humanoid robots are a demand driven product. "A humanoid robot entering the field of health and elderly care, which may integrate various health product functions, such as blood pressure measurement, pulse measurement, or more medical assistance functions. Each person can customize their functions according to their own needs." Wang Zhiwei, Vice President of Data Robotics, said. Manufacturers have various ways to increase people's acceptance of humanoid robots. Wang Zhiwei revealed that the bipedal humanoid robot will be launched to the market by the end of this year, and consumers can modify it according to their own preferences. Next year, a nanny robot will be launched, which will be leased to allow people to experience the service of the robot first. Experts believe that what is unfolding before people is a huge market with a population exceeding that of humans and an industrial scale exceeding 30 trillion US dollars by 2050, and in this field, China will definitely give birth to great enterprises. (Lai Xin She)