Why are humanoid robots so popular?

Author: Du Zhuang

Recently, BYD invested in a humanoid company, Zhiyuan Robotics. This is BYD's first public investment project in the field of humanoid robots.

Not long ago, Zhiyuan Robotics just released the first generation of universal humanoid bipedal robot "Expedition A1". This 175cm tall and 53kg humanoid robot can participate in the appearance process, assemble the chassis, and work on the busy automobile assembly line. It is shown that Shanghai Zhiyuan Xinchuang Technology Co., Ltd. was established in February this year, and its business scope includes robot research and development, theory and software development, artificial intelligence sales, sales, etc.

In addition to BYD, BlueRun Ventures, Wofu Ventures, Qiyu Investment, etc. are also new shareholders of Zhiyuan Robotics in this round. In the six months since its establishment, Zhiyuan Robotics has obtained four rounds of financing, which indirectly confirms the popularity of the humanoid robot market.

In the humanoid robot market, Xiaomi's CyberOne, Dreami's universal humanoid robot, and Cloud Ginger from Dahua have all made their mark. According to Goldman Sachs' forecast, by 2035, the humanoid robot market may reach $154 billion, becoming another landing scenario after smart driving electric vehicles.

Why are humanoid robots so popular? When will they go from being out of reach to being affordable?

What is the difficulty in industrial development?

On May 16 this year, the Opmus humanoid robot was released at the 2023 shareholders' meeting, demonstrating its progress in natural walking, torque control, force control, environmental recognition and memory. Musk emphasized that now is an excellent opportunity to advance the research and development of humanoid robots because software and hardware technology achievements in a large number of fields can be shared. Optimus will use the same fully automatic driving system as the company's electric vehicles. He put forward a bold idea: "A natural person may need to have two humanoid robots. In the future, the number of humanoid robots in the world is expected to reach 10 billion to 20 billion."

Yao Qizhi, academician of the Chinese Academy of Sciences and dean of the Institute of Cross-Disciplinary Information Sciences at Tsinghua University, pointed out that the most ideal body form for embodied general artificial intelligence (AGI) is a humanoid robot.

The purpose of humanoid robots is to interact with artificial tools and the environment, thereby assisting or even replacing human production and life. The technology of traditional robots often focuses on a specific aspect, such as focusing on motion control technology, and sweeping/delivery robots focusing on navigation technology.

Compared with other service robots, humanoid robots require higher perception, motion control, and interaction capabilities. Their motion capabilities represent a system integration project, involving the integration of multiple software and hardware technologies such as autonomous driving, visual navigation, servo systems, reducers, etc. The technology is more integrated and is an extension and evolution of existing applications.

Due to its high complexity and the huge R&D investment required, the development of humanoid robots has become one of the difficulties in robotics technology in the past few decades. For this reason, the pioneers of humanoid robot development, the humanoid robot "Asimo" developed by Honda R&D Institute in Japan, and the humanoid robot "Pepper" jointly developed by SoftBank Group have both been retired.

According to a research report by Minsheng Securities, humanoid robots can be regarded as core components of control systems, bodies, and sensor systems. In terms of voice, vision, decision-making, and control, the combination of large models and humanoid robots helps the latter form a closed loop of perception, decision-making, and control, giving robots common sense. However, the hardware required by humanoid robots, such as servers (similar to human joints), sensors, and ball screws (transmission components), still need to be continuously laid out by the entire industry chain.

In a report on humanoid robots released by Huaan Securities in August this year, it was pointed out that the current humanoid robot industry chain is still in its early stages, focusing on the upstream core components such as reducers, sensors, etc., while the potential application areas downstream are very broad, including logistics, security, elderly care and assistance for the disabled, etc. The report pointed out that if humanoid robots want to be commercialized on a large scale and enter thousands of households, the key is to solve the problem of high costs.

The upstream of the humanoid robot industry chain is basic raw materials and parts, the midstream is the robot body and system integration, and the downstream is different application scenarios. Taking the three core parts of industrial robots, reducers, and controllers, as examples, relevant research reports indicate that the core parts of reference industrial robots account for more than 70% of the cost of industrial robots, of which reducers, servo systems, and controllers account for approximately 36%, 24%, and 12%, respectively. Considering that humanoid robots have greatly improved their freedom compared to traditional industrial robots, it is expected that the cost of reducers and motors will account for a higher proportion.

Why is it so popular?

Many companies have recently launched new humanoid robots. At the same time, more and more universities and research institutions are incubating companies that are turning humanoid robots from laboratories to commercial applications. One of the reasons behind this is the development of large-scale artificial intelligence model technology, which provides technical support for the industrial implementation of humanoid robots. CEOs and other industry insiders believe that the next wave of AI will be embodied intelligence.

At present, the hardware itself has reached a bottleneck period due to the physical limitations of the electric drive form. The real competition is whether the robot is smarter, that is, the artificial intelligence capabilities of the big model. Baidu founder, chairman and CEO Robin Li once said that the big model successfully compressed human cognition of the entire world and allowed us to see the path to achieve general artificial intelligence. "Right now, we are at a new starting point. This is a new era of artificial intelligence with big models at its core. Big models have changed artificial intelligence, and big models are about to change the world."

In June this year, the Ministry released the "China Artificial Intelligence Big Model Map Research Report", which showed that from the distribution of big models released globally, China and the United States are far ahead, exceeding 80% of the global total. The United States has always ranked first in the world in terms of the number of big models.

How can general-purpose robots achieve low-cost mass production and generate practical application value? Peng Zhihui, co-founder of Zhiyuan Robotics, who is known as "Zhihui Jun", said that hardware is only one of the prerequisites, and the more important thing is the robot brain behind it. In Zhiyuan's embodied intelligent brain EI-Brain architecture, the robot's thinking system is divided into the cloud-based super brain, the terminal-side brain, cerebellum, and brainstem, which correspond to the robot's task-level, skill-level, command-level, and servo-level capabilities.

Yao Qizhi said that the emergence of large AI models will create more new value for the industry. To build a general-purpose robot with universal application capabilities, it is necessary to have the three key parts of the human embodied intelligent body, namely the brain, cerebellum, and body, and to make it more adaptable to the human social environment. For embodied intelligent robots, the "brain" dominates the upper-level logical reasoning, decision-making, and planning, and uses natural language to communicate with other intelligent bodies and the environment; the "cerebellum" mainly controls the body by dominating multiple perceptions such as vision and touch, thereby completing complex tasks; and the "body" must have sufficient hardware, such as sensors and actuators.

Currently, many companies have begun to launch exclusive large models for the field of robotics. For example, Google's DeepMind has launched the RT-2 model to improve the robot's natural language understanding ability; domestic player Dahua has released the industry's first large multimodal robot model RobotGPT, etc. This means that the application of general artificial intelligence in humanoid robots will enhance the capabilities of data generation, task decomposition and planning, autonomous operation and control, etc., and has huge application potential.

Dongwu Securities said that with the release of humanoid robot products by Tesla last year, the commercialization of humanoid robots has accelerated. Under the leadership of Tesla's humanoid robot benchmark, it is expected that the global humanoid robot will be initially commercialized in 2025, with sales of 20,000 to 30,000 units. By 2030, there will be an additional demand of 1 million units, and sales are expected to exceed 10 million units in 2035.

Approaching the critical point of transformation

At present, the robotics industry is gradually approaching the critical point of transformation and leap. The "China Robotics Technology and Industry Development Report (2023)" points out that the commonality and cutting-edge technologies of robots are developing rapidly, and a large number of science and technologies such as,, biology, and materials are promoting each other in mutual integration. Integration of innovation, high-end, intelligent, and ecological construction have become the keywords for the future development of Chinese robots.