The human imagination and contemplation of artificial machinery have a long and storied history.
In Greek mythology, there are mechanical beings and artificial humans such as Hephaestus's golden robots and Pygmalion's Galatea. According to the "Liezi ยท Tangwen" compiled by Liezi, stories of artificial people were already present during the Western Zhou period in China.
On July 4th, when the 2024 World Artificial Intelligence Conference opened in Shanghai, the shock of presenting 18 humanoid robots before the exhibition was indescribable. The event showcased 45 intelligent robots, including 25 humanoid robots.
01
The Birth of Humanoid Creatures
Tesla's Second-Generation Humanoid Robot, Optimus
At the scene, there was a bustling crowd. The second-generation Optimus is equipped with AI technology that is homologous to Tesla vehicles. Based on the ability to walk upright, the walking speed of the second-generation Optimus has increased by 30%; its fingers have also "evolved" to the point where they can not only sense and touch but also handle delicate tasks such as holding an egg gently and carrying heavy objects with ease.
Tesla expects to start limited production of humanoid robots next year, with more than 1,000 Optimus robots helping humans complete production tasks in Tesla factories.Open Source General-Purpose Humanoid Robot "Qinglong"
Humanoid Robot (Shanghai) Co., Ltd. released China's first full-size open source general-purpose humanoid robot on the same day. This robot is 185cm tall and weighs 80kg, with a highly biomimetic torso structure and anthropomorphic motion control. It supports multimodal mobility, multimodal perception, multimodal interaction, and multimodal manipulation, with up to 43 active degrees of freedom throughout the body, a maximum joint torque of 400N.m, and computing power support of 400TOPs. It can walk at a speed of 1m/s while carrying a load of 40kg.
Leju "Kuafu" (KUAVO) Robot
"Kuafu" (KUAVO) is the first full-size humanoid robot in China equipped with the HarmonyOS operating system, and it is also the first full-size humanoid robot in China to achieve industrial landing, also showcased at the exhibition booth.
Kuafu was released in December 2023, weighing about 45kg, with 26 degrees of freedom throughout the body, a maximum walking speed of up to 4.6km/h, capable of rapid continuous jumping, and a jumping height exceeding 20cm. It is equipped with self-developed integrated joints and a depth camera, enabling all-around visual perception.
Currently, the Kuafu robot has achieved generalized operation of humanoid robots under small samples, and typical transaction cases include scientific research cooperation with Shandong University and the Beijing General Artificial Intelligence Research Institute.
Tianlian Humanoid Robot T1
(Note: The original text was cut off, so the translation for "Tianlian Humanoid Robot T1" is incomplete.)The humanoid robot stands approximately 1.60 meters tall, with a total of 71 degrees of freedom. The bare machine of the humanoid robot weighs only about 37 kilograms, and with the battery, it is around 43 kilograms, capable of performing a weighted squat that exceeds 145 kilograms.
The staff introduced that this is the highest level known in the industry at present. It can perform movements such as splits, jumps, spins, and kicks, with flexibility comparable to that of a dancer. It is worth mentioning that the Tian Lian humanoid robot T1 is entirely self-produced and self-developed throughout the entire industry chain.
Unitree Technology Universal Humanoid Robot H1
This is the world's first universal humanoid robot with the capability of performing a backflip in place. The realization of such a high-difficulty movement not only demonstrates the robot's excellent performance in balance control and power systems but also indicates that humanoid robots may play an important role in complex terrains and extreme environments in the future. Currently, the "Unitree H1" robot has achieved functional applications in industrial scenarios.
Unitree Technology CEO Wang Xingxing said, "Embodied intelligence is the most effective way to achieve AGI (Artificial General Intelligence), and the general direction of OpenAI is likely wrong." He believes that there will definitely be significant technological breakthroughs in the next 5-10 years, and only by believing in AI as much as possible will there be a better future.
Kepler Universal Humanoid Robot K1
This robot stands 175 cm tall and weighs 70 kg, with a total of 40 joint degrees of freedom and more than 80 sensors throughout the body. It is also equipped with China Telecom's Starry Sky large model, giving it various capabilities such as walking on complex terrain and intelligent obstacle avoidance.
"We already have the potential for commercial application, and we plan to start small-batch production in October this year," said Kepler CEO Hu Debo. He indicated that there are many intent orders from industries such as logistics, automobile manufacturing, and metallurgy, "The pace of commercialization of humanoid robots is faster than expected."How much do humanoid robots cost?
Many exhibitors have indicated that there is still a distance before humanoid robots can enter the consumer market. In this gap, cost is an issue. For consumers, the current price of humanoid robots is too expensive.
For example, the price of Tesla's second-generation humanoid robot Optimus is around $20,000 (140,000 RMB), Data Robotics' price starts from 399,000 RMB, Tianlian Robotics' price is around 500,000 to 1,000,000 RMB, "Kuafu" (KUAVO) robot's price is between 600,000 to 800,000 RMB, and the unit price of Unitree H1 is about 650,000 RMB.
Apart from Tesla, which has clearly stated that it will sell to individuals in the future, the quotes for other humanoid robots are all in the hundreds of thousands, and if equipped with a complete "brain," "cerebellum," "flexible hands," and "lidar," the price can easily reach millions. The high cost is not consumer-friendly.
On the other hand, there is still a safety issue. During on-site exchanges, most manufacturers expressed some concerns about "safety." For example, in non-close contact scenarios, people can maintain a certain distance from the robot and observe its operation while ensuring their own safety. However, in scenarios that require physical contact, such as elderly care or personnel assistance, the safety and risk issues of humanoid robots cannot be ignored.
Looking at the application of humanoid robots, most applications are still in a point-to-point mode, and the current application scenarios are mainly: scientific research, intelligent inspection, intelligent manufacturing, warehousing and logistics, outdoor operations, etc.
Industry will be the mainstream scenario for the first large-scale application of humanoid robots.
Zeng Xuezhong, Senior Vice President of Xiaomi Group, predicts that in the future, intelligent factories will be composed of three types of labor: 70% automated equipment, 20% humanoid robots, and 10% humans. Humans will be responsible for value judgment and mining, while humanoid robots will be responsible for achieving flexible manufacturing and cross-system collaboration. The collaborative work of the three types of labor will redefine the concepts of "workers" and "factories."
Therefore, the implementation of humanoid robots will start by entering a scene (such as industry), achieving multiple functions, expanding into multiple scenes (logistics, agriculture, consumption, etc.), achieving complex functions, and then gradually entering thousands of households.Discussion on Embodied Intelligence
In the slogans of every exhibitor, they are all claiming to be "embodied intelligence," the "embodiment of AI," and the "new species" of the future.
Here, I would like to take some time to discuss the concept of "embodied intelligence."
Last year, the Beijing Municipal Science and Technology Commission, Zhongguancun Management Committee, and other departments jointly issued the "Beijing Measures to Promote the Innovative Development of General Artificial Intelligence (2023-2025) (Draft for Comments)," which proposed exploring new paths of general artificial intelligence such as embodied intelligence, general intelligent agents, and brain-like intelligence, and promoting research and application of embodied intelligent systems, breaking through the perception, cognition, and decision-making technologies of robots in complex conditions such as open environments, general scenarios, and continuous tasks.
What is embodied intelligence? The MBA Wisdom Encyclopedia defines it as creating intelligent agents that combine software and hardware, which can be simply understood as various forms of robots, allowing them to perform various tasks in the real physical environment to complete the evolution process of artificial intelligence.
In fact, the difference between embodied intelligence and GPT-4 is that GPT-4 and others are passively learning, that is, the machine learns what kind of data (human labels or generated data) we provide to it. Embodied intelligence is a new form of autonomous learning, capable of self-perception and understanding of the physical world, just like human ability to perceive and understand the external environment.
In summary, embodied is an intelligent agent with a body, in English it is called Embodied AI, which means giving an intelligent agent a body, and this is embodied intelligence.
With the progress of large language models in the past two years, we have seen that general intelligence in natural language can be achieved. The field of computer vision has also made great progress in 3D understanding and open semantic recognition, and the industry is looking forward to the breakthrough of general embodied intelligence.
At the World Artificial Intelligence Conference, Yu Shi Technology CEO Wang Xingxing said that for embodied intelligence, it is very important to have a world model that can understand time, space, and physical laws. "The current large language models are very much like living in a dream," AGI needs to have physical interaction with real robots to deepen imitation and reinforcement learning, and it also needs to participate in human life to experience and understand human emotions and personality.From a future perspective, embodied intelligence will involve an intelligent brain working in conjunction with various body configurations to accomplish different tasks in the physical world. Consequently, the industry has been discussing an intriguing question: Must embodied intelligence be humanoid? Must it be bipedal? Can embodied intelligence be achieved through robotic dogs?
04
Humanoid Robot Investment Boom
In the first half of 2024, the financing heat for the humanoid robot track continued to rise.
According to incomplete statistics, there were more than 22 financing events in the global humanoid robot field in the first half of 2024, with the total financing amount exceeding 7 billion yuan. The American humanoid robot startup Figure AI led the world with a huge financing of 675 million US dollars (equivalent to about 4.9 billion yuan in RMB), while Yushu Technology topped the domestic Chinese humanoid robot financing list with nearly 1 billion yuan in financing.
In the A-share market, the performance of humanoid robot concept stocks is also eye-catching. On July 4th, humanoid robot concept stocks soared after the opening. Wind data shows that Henggong Precision quickly achieved a 20% limit up after the opening, followed by Lei Sai Intelligence, Shiyun Circuit, and others also hitting the limit up, with Siling Shares, Buk Technology, Beit Technology, Jiangsu Leili, and others following suit.
However, many industry insiders at the on-site exchange have unanimously expressed: If robots are to be widely used in the future, from manufacturing to services, and truly enter public life, achieving tasks such as accompanying the elderly and teaching children, there are still many "hardware and software" that need to be upgraded. This includes visible parts such as the sensitivity and safety of components, as well as invisible aspects such as data privacy, algorithmic bias, and ethical issues, all of which require more comprehensive and optimized solutions to ensure the safety and sustainable development of new technologies and products.
As Turing wrote in "Computing Machinery and Intelligence" in 1950: "The view is only a short distance ahead, but even so, there is still a lot of work worth doing waiting for us there."
Comment