In recent times, the combination of large-scale models and robots has been attracting a lot of attention, from the proposal of "embodied intelligence" by Professor Fei-Fei Li's team at Stanford University, to the recent "pride of domestic production" by brands such as Yushu and Zhiyuan. At the 2023 World Robot Conference a few days ago, humanoid robots performed various skills on the spot, bringing science fiction into reality. Various enterprises showcased their robot-related products at the event. After the conference, it was found that a total of 160 domestic and foreign robot companies and 600 robots participated, with humanoid robots taking the spotlight. Renowned Boston Dynamics founder Marc Raibert and Japanese robot scientist Hiroshi Ishiguro also attended the conference.
Undoubtedly, humanoid robots are becoming a new hot spot in the capital market. In the primary market, investors such as Baidu Ventures, Wei Capital, Hillhouse, GGV Capital, and ZhenFund are actively conducting research. In the secondary market, the concept stocks of humanoid robots have experienced several waves of speculation. In May, Musk's remarks directly led to a surge in the A-share robot concept stock "CIMC Intelligent" and "Fengli Intelligent" also rose by over 150% within six trading days, even attracting the attention of the Shenzhen Stock Exchange, which requested an explanation for the large increase in stock prices and its reasonableness.
Tesla's Drive
At this year's Tesla shareholder meeting, Musk said, "The humanoid robot Optimus has significantly enhanced control over motion and strength, as well as environmental perception, and the technology is iterating rapidly. It is expected that the demand for robots in the future may reach 10 billion, or even more. If we calculate based on a ratio of 2:1 between humans and robots, the demand for humanoid robots may be even greater than that of electric vehicles." With just a few words, Musk's confidence, enthusiasm, and dedication quickly ignited the flames of the humanoid robot race, and the trend has been growing stronger ever since. From the perspective of practical application, this may also drive the entire industry chain towards maturity and perfection, which is also likely to be the case for Tesla.
Tesla's humanoid robot Optimus made its first appearance at the official 2022 AI Day event last year and completed autonomous walking, turning, stopping, and waving gestures on the spot. Most of Optimus's technology is consistent with Tesla's vehicles, such as machine vision, decision-making based on visual data, support for communication, and the "brain," as well as the most important chip consistent with Tesla vehicles. It also incorporates the FSD computer and Autopilot-related neural network technology, with a final expected price of no more than $20,000, or about 144,000 RMB.
From a practical technological perspective, companies like Tesla have inherent advantages in making robots because many of the basic principles of robots and AI are the same, and it can also be seen as a natural extension of electric vehicles, which can be seen as the first generation of four-wheeled robots. When Musk proposed to make robots a few years ago, he attracted a lot of ridicule for "not focusing on the main business." However, at the beginning of this year, Ideal Car confirmed its vision to become the best AI and robot company, rather than the world's largest electric vehicle company. This shows that dreams can also spread.
New Directions for Robots
Robots are now commonplace, and what determines how far humanoid robots can go is their task generalization ability, which many companies are currently focusing on. The "pride of domestic production" such as Yushu and Zhiyuan, as mentioned earlier, will not be elaborated here. Other startups are also worth paying attention to, such as Yuequan Bionics, whose robot products are as dexterous as human hands. The company's independently developed humanoid bionic intelligent dexterous hand can adjust its movements, such as active movement, bending, and flipping, to prevent objects from falling when subjected to external forces. The company states that, "In addition to basic gripping and pressing actions, it can also perform 27 different complex and delicate hand operations, such as using chopsticks to pick up small objects, applying skincare products, stirring coffee, brushing a phone, and unbuttoning clothes." The dexterous hand uses a tension-compression body drive technology and has a high degree of freedom. It also incorporates built-in flexible sensors with tactile feedback.
In addition, there is the "Wukong-4" humanoid robot developed by the Robotics Team at the Zhejiang University School of Control Science and Engineering. According to the introduction, "Wukong-4" can adapt to various terrains such as outdoor roads, grass, and muddy surfaces, with a maximum forward speed of 6 kilometers per hour, the ability to jump 0.5 meters high, and the capability to traverse slopes of up to 25 degrees and 10-centimeter-high steps. In the face of unknown disturbances such as slippery road surfaces and external thrust, "Wukong-4" can quickly regain balance and maintain stable walking. By integrating leg-foot motion technology with environmental perception technology, "Wukong-4" has achieved three-dimensional environmental map construction and autonomous dynamic navigation for robots.
Behind these new robotic products are a series of software and hardware technologies developed independently by enterprises and universities: humanoid robot body based on self-perception drivers; the use of self-developed high-torque density modular joints and integrated structural design; advanced materials such as high-strength alloys, carbon fiber, and engineering plastics to retain an aesthetic appearance, improve structural strength, and stability; coupled with large language models and advanced force control algorithms, they possess high dynamic performance while better understanding human beings.
From the pursuit of investors, the entry of leading technology companies, to the innovative research of startups and universities, humanoid robots are making progress in multiple directions, ushering in a new stage of development. For example, integrating large-scale models and equipping them with a brain is another key technological variable driving the humanoid robot craze.
It was previously mentioned that "embodied intelligence" refers to "intelligent entities with actual bodies that support physical interaction," which is equivalent to adding a body to AI and represents a new development path for general-purpose robots. Capital is adding fuel to the industry, and after five to ten years of large-scale investment, it will make it burn brighter and ultimately achieve the commercialization of general-purpose robots. For a long time, the development of robots has been limited to completing specific tasks, just as AI was once limited. Large-scale models are gradually evolving into AGI, and their combination with robots can naturally expand the application scope of robots.
Compared to domestic progress, foreign progress is one step ahead, having already upgraded the capabilities of large-scale models from language to execution. In July, Fei-Fei Li's team's robot was able to open drawers, twist bottle caps, and weigh apples. At the end of the month, Google DeepMind introduced the Robotics Transformer 2 (RT-2), continuing to delve into the same direction. RT-2 is a new visual + language + action (VLA) model that can learn from networks and robot data and transform this knowledge into general instructions for robot control. RT-2 has demonstrated better generalization capabilities, understanding semantics and visual categories beyond the robot data it has encountered, and can interpret new instructions and respond to commands through basic reasoning.
Destined for a Rocky Road
As the saying goes, "Ideals are full, reality is bony." While robots and AI are accelerating their development, there are still many real technical challenges and commercialization challenges. Some believe that the field of general-purpose robots also needs a technology or product like GPT-4 to combine multi-modal capabilities and truly unify the development of embodied intelligence. However, this is no easy task. Currently, the robots demonstrated in papers and some demos that are combined with large-scale models tend to focus on solving interaction problems, but solving interaction problems does not automatically turn robots into general-purpose robots. From the development history of robots, it is clear that large-scale models have made contributions, but their influence on underlying control and execution is limited. Currently, the academic community generally hopes to use AI-driven methods to implement reinforcement learning as the underlying control, but this is not directly related to large-scale models, and the control methods of reinforcement learning are also immature, with most still in the academic research stage.
Another challenge lies in the coordinated evolution of software and hardware capabilities. Although many people advocate that the combination of large-scale models and robots will bring disruptive opportunities, Boston Dynamics founder Marc Raibert has stated that in the future development of robot technology, hardware engineering and software are equally important. "Some people believe that software can overcome all problems and limitations in hardware, but I do not agree with this view. Only the best hardware designers and software designers working together can design the best robots in the world."
Furthermore, there is a need for improvement in safety. The "AI illusion" of large-scale models may not have any substantial impact, but once robots enter daily life, they need to ensure accuracy and safety, which are areas where technology needs to improve. Technology, scenarios, costs, safety, opportunities, and challenges are all arriving simultaneously, and humanoid robots are taking a crucial step towards the future.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
