In recent years, artificial intelligence has seen remarkable advancements and innovations, positioning itself to profoundly shape the future of technologyOne of the most significant emerging trends is embodied intelligence, or embodied artificial intelligence (EAI), which refers to intelligent systems that can perceive, reason, and engage with the physical worldThis concept is not just theoreticalNVIDIA’s CEO Jensen Huang forecasted last year during the ITF World 2023 Semiconductor Conference that artificial intelligence's next wave would focus on such embodied capabilities, and events over the past months have confirmed that this prediction is on the verge of becoming a reality.

In May, at the International Conference on Robotics and Automation 2024 held in Yokohama, Japan, humanoid robots took center stage, showcasing their potential in advanced roboticsThe fascination with humanoid robots continued in July at the World Artificial Intelligence Conference 2024 in Shanghai, where the star attraction was the humanoid robot known as the "Eighteen Mighty Ones", sparking widespread interest and admiration

This enthusiasm culminated in August during the World Robot Conference in Beijing, where humanoid robots were also prominently featured, with officials announcing that this edition featured the highest number of humanoid robots ever, captivating over half the audience's attention at the humanoid robot company booths.

Clearly, humanoid robots are stepping into the limelight, much like large language models have doneThese robots, designed with advanced capabilities, are not just machines; they have been programmed to perform a wide array of tasks, including household chores, writing, and even providing martial arts training or personal securityHowever, beneath the surface excitement, a fundamental question remains: how far are these humanoid robots from truly being able to "make a living" through their capabilities?

The surge in interest in robotics can be traced back to developments in large AI models

• Indian Rupee Expected to Devalue by 3% in 2024
• The Truth About the Silicon Carbide Market
• Trade with India Has Quadrupled Since 2005
• Collapse of Europe's Semiconductor Industry
• Fed Rate Cut May Be Delayed

Before diving deeper, it is important to clarify the concept of embodied intelligenceThis form of intelligence incorporates two essential components: "embodiment" and "intelligence." EAI systems possess characteristics such as perception, decision-making, interaction with the environment, and physical presenceEssentially, embodied intelligence allows machines to discern their surroundings and execute specific tasks in the real world.

As early as six decades ago, tech pioneers like Omron proposed the vision that machines should handle tasks they are capable of, thus freeing humans for more creative endeavorsHowever, limitations in technology stifled the full realization of this vision until recentlyThe resurgence of embodied intelligence can be attributed largely to the advancement of large AI models, which began to flourish in 2022.

An example of this evolution is the humanoid robot Figure02, launched by startup Figure AI, which utilizes a collaborative AI model developed in partnership with OpenAI to enhance real-time interaction and common-sense reasoning capabilities

Such advancements highlight that large model technology provides a crucial foundation for humanoid robots, enabling them to better perceive, decide, and interact with their environmentAdditionally, hardware advancements have also significantly progressed, as evidenced by Tesla’s release of the Optimus Gen2 robot equipped with self-developed actuators, which has improved its walking speed and stability.

This buzz has also had a remarkable impact on the market; after Tesla announced Optimus, the company’s stock rose for eleven consecutive trading days, adding around $67 billion to Elon Musk’s net worthThis further validates the commercial potential of humanoid robots, pushing the focus of technological investment back toward embodied intelligence.

Three key aspects illustrate this phenomenon: firstly, governments worldwide are guiding policy development; in China, short-term goals focus on technological breakthroughs in core components, while long-term goals emphasize industry applications and ecosystem expansion

In contrast, many foreign policies target the advancement of cutting-edge technologies and the implementation of significant applications.

Secondly, the field is becoming increasingly competitive, with numerous players entering the marketAlongside established companies like UBTech, Boston Dynamics, and Fourier Intelligence, numerous startups such as Zhiyuan Robotics and Galaxy General Robotics are emergingAdditionally, tech giants like iFlytek, Baidu, Tencent, and even companies outside the traditional robotics field, like Xpeng Motors and Xiaomi, are joining the race, often equipped with a clear vision of applicable scenarios and sharing smart manufacturing supply chains.

Thirdly, investment in embodied intelligence is growing, driven by the quest for returns on substantial financial commitmentsRecent statistics reveal that nine humanoid robot companies in China raised over 1.9 billion yuan in funding throughout 2023 alone, while 13 others secured more than 2.5 billion yuan in the first half of the year.

Despite this burgeoning excitement, many challenges still face the humanoid robotics industry, particularly at the research and application levels

A critical concern is the convergence of technical pathways, as articulated by Wang Tianmiao, honorary director of the Robotics Research Institute at Beihang UniversityHe mentioned two significant challenges: the soft-ware side remains in development, with suitable general and specialized models for robots still under exploration, while the technology for dexterous hands is still costly and complex.

The software aspect refers to the capability of breaking down complex tasks into multiple smaller tasks, requiring the facilitation of large model technologies for effective human-robot interactionThis generalization capability is crucial for a service robot, for example, which should autonomously prioritize its chores without needing explicit commands from its owner.

While hardware might not pose significant barriers, differing mechanical capabilities among robots highlight the ongoing need for advanced software to define their effectiveness

The fundamental requirement is to establish a generalized capacity within humanoid robots so they can adapt to various task scenarios, ensuring their usability.

When discussing the humanoid robotics landscape, further divergences emerge in the form of design choices related to the robots’ “hands” and “feet.” For movement capabilities, industry factions are divided between bipedal and non-bipedal configurationsThe former argues that, mirroring traditional humanoid designs, bipedal robots reflect the ultimate goal of embodied intelligenceHowever, developing usable, stable, and cost-effective bipedal robots remains an uphill battle compared to non-bipedal alternatives, which have shown greater adaptability in commercial applications.

With regards to robotic dexterity, the variety of design decisions is even greater, ranging from companies opting to create humanoid robots equipped with five fingers—like the Sparky1 by Daimeng Robotics, which has optical tactile sensors—to those choosing to start with simpler designs such as two or three-fingered setups, as seen in the Galaxy General G1 or UniX AI's Wanda

The diversity of choices can be partly attributed to competing manufacturers defining their own technological standards before any consensus has been achieved.

A more convoluted challenge for humanoid robot manufacturers, however, lies in data collection for training purposesEnhancing a robot's software capabilities requires extensive training data, which often necessitates gathering human behavioral data from various job tasks through added sensorsThis complexity arises when actual human involvement in data collection is required in practical settingsFor instance, Zhiyuan Robotics has announced its plan to build a sampling facility with about 100 robots and 150 workers to gather production data effectively.

As humanoid robots gradually enter small-scale production, many of these companies are also striving to set competitive pricing modelsExamples include the Yushi Technology G1 humanoid robot, which was priced below 100,000 yuan and generated significant interest upon its launch

Meanwhile, Zhiyuan plans to commence mass production this October, expecting to produce approximately 100 units each month.

While this news is promising, the journey to viable commercial opportunities for humanoid robots is still fraught with difficultiesExperts note that many currently available models primarily serve as demonstration tools for research platforms, equating them to operational concepts in the autonomous vehicle spaceEven with loads of enthusiasm surrounding this sector, it is essential to emphasize that the actual pathways to commercial success may still be in the infancy stage, with three primary avenues of opportunity echoing those found in the autonomous driving realm.

One opportunity involves focusing on local development similar to L4 autonomous vehicles, which could put companies in a prime position once the era of embodied intelligence fully arrives

The second opportunity lies in specific applications, similar to mining and sanitation scenarios currently being explored, although outcomes are still unclearLastly, upstream and downstream industrial opportunities remain lucrative, encompassing everything from building intelligent computing centers to key technologies in sensors and actuator modulesThese segments often yield higher profits than the core products themselves.

Despite varied opinions on the market share of humanoid robots, which some optimistically predict could exceed 60%, others point toward a bleaker scenario where they might only capture 30% due to vastly diverse applications of other robot typesHence, the final shape of market competition will depend heavily on application contexts, client needs, and the technology's evolving abilities.

Last but not least, according to predictions from the International Federation of Robotics, the global humanoid robot market is expected to see an astonishing compound annual growth rate of 71% between 2021 and 2030, and similar projections in China anticipate a market scale of approximately 870 billion yuan by 2030.