China's humanoid robot market to hit 10 billion yuan

• GLOBAL TIMES

• Discover China, Discover the World

Download App

China's humanoid robot market to hit 10 billion yuan, eyes 119 billion yuan by 2030: report

Leng Shumei01:16 Apr 12 2024

Photo: VCG

Research unveiled during the inaugural China Humanoid Robot Industry Conference in Beijing from April 9 to 10 indicates that the global humanoid robot industry is entering a golden era, poised for sustained growth. The report predicted that the Chinese humanoid robot market will surpass 10 billion yuan, reaching 10.47 billion yuan ($1.45 billion) by 2026, and is anticipated to soar to 119 billion yuan by 2030.

The booming market is seemingly telling people that intelligent humanoid robots that can simulate human thinking and consciousness, as depicted in films like Ex Machina and A.I. Artificial Intelligence, are really getting closer to reality.

The rapid development of AI technology has played a crucial role in the advancement of humanoid robots. As NVIDIA CEO Jensen Huang said at the 2024 GPU Technology Conference (GTC) in March, "Building foundation models for general humanoid robots is one of the most exciting problems to solve in AI today."

In return, the advancement of humanoid robots is viewed as a major milestone in the AI era, pushing the boundaries of AI research. Numerous Chinese experts and industry observers consider humanoid robots a breakthrough for the "AI Plus" initiative aimed at fostering innovative development in the digital economy, as promoted during the two sessions in March.

Golden age along with AI

Humanoid robots are dubbed "humanoid" because they are designed to emulate and potentially surpass human capabilities in form, function, behavior, and even cognitive processes, Zhang Rui, founder and executive director of the Beijing Ironman Technology in Beijing, told the Global Times.

"Without the need for massive changes to the existing environment, humanoid robots can seamlessly integrate into various scenarios, using their flexible and dynamic execution capabilities to meet complex and changing task requirements. Furthermore, their human-like characteristics enable them to easily manipulate human tools, further expanding their application areas," Zhang said.

Therefore, humanoid robots are not only a symbol of technological progress but also a significant force driving future social development, he noted.

Humanoid robots have been widely applied in various industries, with the aerospace sector being one of the most prominent, according to Zhang. Several countries including the US, Russia and China, have been deeply researching the application of humanoid robots in the aerospace field. These robots are mainly used to replace humans in performing dangerous and complex operations, ensuring the safety of astronauts and improving the efficiency and success rate of space missions.

Other major application areas of humanoid robots is in border defense and lights-out factories, or smart factories.

"The continuous innovation and breakthroughs in AI technology in recent years have indeed provided humanoid robots with more powerful perception, decision-making, and execution capabilities. This allows humanoid robots to more accurately understand human language, recognize environmental information, and make more reasonable decisions and actions," Zhang said.

In the future, Zhang expects humanoid robots to have enormous potential in areas such as general hardware execution, dynamic adaptation and environmental integration.

How far is AGI?

The thriving progress of humanoid robots is drawing increasing attention from international tech giants. On March 18, nine humanoid robots were unveiled at NVIDIA's 2024 GTC. Tesla is also actively working on a humanoid robot named Optimus, and OpenAI, Microsoft, and Amazon founder Jeff Bezos have made substantial investments in humanoid robot startup Figure AI. Additionally, Agility Robotics, backed by Amazon, has established the world's first large-scale humanoid robot production factory in Oregon, the US, capable of producing 10,000 two-legged robots annually.

A group of innovative and competitive Chinese companies have also emerged in the field of humanoid robots with the increasing emphasis and investment in robot technology in China, leading to significant progress in the Chinese humanoid robot industry.

In March, the Beijing humanoid robot innovation center announced that it would soon release the first generation of a universal open humanoid robot body.

Among the nine robots showcased at the NVIDIA 2024 GTC, two were developed by Chinese companies, namely H1 from Hangzhou Unitree Robotics and PX5 from Xiaopeng Pengxing.

The Global Times learned from Unitree Robotics that H1 is a full-size humanoid robot capable of running, equipped with 360 panoramic depth perception. Currently, it can reach a speed of 3.3 meters per second, setting a world record for full-size electric humanoid robots, with a potential speed of up to 5 meters per second.

This robot boasts highly advanced full-body dynamic coordination capabilities, enabling it to dance in groups and execute backflips. As a result, NVIDIA opted to partner with Yushu to collectively propel the global advancement of AI robots. According to the company's response to the Global Times, NVIDIA, a frontrunner in GPU and AI chip technology, furnishes Yushu's robots with robust computing capabilities and comprehensive support in deep learning technology.

However, overall, the deep integration of AI and humanoid robots still faces significant challenges.

Zhang believes hardware challenges are a crucial obstacle. While we can achieve various complex functions and performance at the algorithm level, it is often difficult to achieve the desired output power and efficiency in actual robot hardware, he said.

This is mainly due to the numerous technical details and engineering challenges involved in hardware design and manufacturing, requiring continuous optimization and improvement, Zhang explained.

On the other hand, the current progress of AI technology is mainly limited to deepening and innovating at the logical level, with insufficient breakthroughs in thinking and emotional aspects. While the form of robots is malleable, the "spirit" of their internal thinking and emotions is still an unexplored frontier. Zhang believes it will take another 5-10 years to achieve a 70 percent similarity with human emotions.

At an economic forum held at Stanford University in March, Jensen Huang predicted that a general artificial intelligence that can pass human tests, or "human-like" artificial general intelligence (AGI, capable of performing all human intelligent behaviors), is likely to appear within five years. However, Huang also pointed out that achieving this goal is not without difficulties, as scientists still lack a unified definition of how the human mind operates, making it challenging for engineers to achieve the goal.

On April 8, Elon Musk said during a livestreamed interview on X that AI that is smarter than any one human will probably come around by the end of next year. Last year, he predicted that humans would "fully" achieve general artificial intelligence by 2029.

Some experts believe that with the continuous advancement of chips and algorithms, AI may eventually surpass human intelligence. However, Liu Wei, director of the human-machine interaction and cognitive engineering laboratory of the Beijing University of Posts and Telecommunications, pointed out that AGI may be a false proposition. This is not because current AI systems have not reached the level of general intelligence, but because AI fundamentally perform and learn like humans.

The development of AGI faces three major bottlenecks: technical, biological, and social. The technical bottleneck lies in the need for AI systems to have higher computing power, more advanced algorithms, and more efficient data processing methods to achieve more complex and intelligent functions. The biological bottleneck mainly stems from our limited understanding of the cognitive capabilities and operation mechanisms of the human brain, requiring deeper research in neuroscience and cognition to achieve similar levels of intelligence. The social bottleneck includes the integration of AI systems with human society, such as cultural differences, ethical issues, privacy protection, etc., all of which are crucial factors affecting the development of AI, according to Liu.

"To overcome these bottlenecks, interdisciplinary cooperation and continuous innovation efforts are needed. Only by making breakthroughs in technology, biology and society can AGI move towards more mature and comprehensive development, but it remains extremely difficult, perhaps impossible after all," Liu said.

Safety and ethics concerns

In November 2023, the Ministry of Industry and Information Technology of China issued the guiding opinions on the innovation and development of humanoid robots, proposing to establish a preliminary innovation system for humanoid robots by 2025.

The document predicted that, by 2025, key technologies such as those related to robots' "brain, cerebellum, limbs" will achieve breakthroughs, ensuring safe and effective supply of core components. The whole machine products will reach international advanced levels, achieve mass production, and be demonstrated in special, manufacturing, and civil service scenarios, exploring effective governance mechanisms and means. By 2027, China's comprehensive strength in humanoid robots will reach world-class levels, becoming an important new engine for economic growth.

The development of the humanoid robot industry, driven by both technology and policy, has entered the fast lane of development, but it also faces challenges in safety, social ethics, and legal norms. Some experts point out that in scenarios requiring close contact, such as in elderly care or assistance, where physical contact is needed, the safety and risk issues of humanoid robots cannot be underestimated. More research is needed by the industry and relevant regulatory authorities on the technical and application scenarios.

Zhang believes that when unexpected situations occur with humanoid robots in a home environment, the lack of clear legal definitions makes it difficult to determine responsibility and protect rights. Additionally, there is currently no mandatory system for certifying the eligibility of humanoid robot products, making it challenging to distribute products in large quantities and limiting the widespread application of humanoid robots. Therefore, it is essential to address and resolve these legal gaps as soon as possible to ensure the healthy and orderly advancement of humanoid robots in the future.

However, he firmly believes that with the continuous progress of technology and the gradual improvement of regulations, humanoid robots will demonstrate their unique charm in more fields, contributing more to the development of human society.

In 2023, the Chinese Ministry of Science and Technology, together with the Ministry of Education, the Ministry of Industry and Information Technology, and 10 other departments, issued the trial ethical review measures for science and technology.

According to the measures, carrying out scientific and technological activities should adhere to the unity of promoting innovation and preventing risks. It involves objectively evaluating and prudently handling uncertainties and risks of technological applications, following the principles of enhancing human well-being, respecting the rights of life, upholding fairness and justice, reasonably controlling risks, and maintaining openness and transparency in science and technology ethics. It is essential to comply with the Constitution, laws, and regulations of China, relevant provisions, and ethical norms of science and technology.

AI助推发展，智能人形机器人还有多远

文/冷舒眉

在科幻作品中高端的人工智能（AI）往往以人形机器人为载体，而在现实中，随着人工智能技术的突飞猛进，如《机械姬》《人工智能》等影视作品中畅想的可以模拟人类思维和意识的智能人形机器人，似乎也离我们越来越近。

根据于4月9日在首届中国人形机器人产业大会上发布的《人形机器人产业研究报告》，全球人形机器人行业发展正在进入黄金时期，市场规模将呈现持续增长态势。预计2026年中国人形机器人市场规模超百亿，达到104.71亿元人民币，2030年有望成长为千亿市场，市场规模达1192.46亿元。

近几年AI技术的快速发展无疑对人形机器人的进步起到了至关重要的作用。同时，人形机器人的发展也被看作是人工智能时代的重要标志之一，在某种程度上推动着人工智能的前沿研究。正如英伟达CEO黄仁勋在2024GPU技术大会（GTC）上所说“开发通用人形机器人基础模型是当今AI领域中最令人兴奋的课题之一”。那么智能人形机器人离我们还有多远？

AI带来发展黄金期

人形机器人之所以名之以‘人形’，旨在追求与人类相仿的形态、行为乃至思维模式的模拟，北京钢铁侠科技有限公司创始人张锐对《环球时报》如此说。

“人形机器人，作为人类智慧的结晶，其形态与功能的设计旨在模仿并超越人类的能力，从而成为未来社会中不可或缺的通用型智能执行者。在无需大规模改变现有环境的前提下，人形机器人便能无缝融入各类场景，以其灵活多变的动态执行能力应对复杂多变的任务需求。更值得一提的是，其仿人的特性使得人形机器人能够轻松驾驭人类工具，进一步拓宽了其应用领域。因此，人形机器人不仅是科技进步的象征，更是推动未来社会发展的重要力量。”

目前人形机器人已经被广泛应用到了各个行业中，张锐认为航天领域是其中最为突出的一个。目前全球范围内，包括美国、俄罗斯、中国等国家都在深入研发人形机器人在航天领域的应用。这些机器人主要用于替代人类完成一些危险、复杂的进出仓操作，从而确保航天员的安全，并提高航天任务的效率和成功率。

除此之外，边海防、面向危险作业的“人机黑”（人机隔离、机器换人、黑灯工厂），以及智能制造等也是当前人形机器人的主要应用领域。

“近年来人工智能技术的不断创新和突破，为人形机器人提供了更加强大的感知、决策和执行能力。这使得人形机器人能够更准确地理解人类语言、识别环境信息，并做出更加合理的决策和行动，”张锐说。未来，人形机器人的研发在在通用硬件执行、动态适应、环境融合、工具利用以及广泛通用性等方面都拥有巨大潜力，张锐说。

智能人形机器人还有多远

在AI助推下蓬勃的人形机器人正在吸引越来越多国际科技巨头的关注。美国时间3月18日，九款人形机器人在英伟达2024GPU技术大会（GTC）上压轴登场。特斯拉也在紧锣密鼓研发人形机器人“擎天柱”，OpenAI、微软、亚马逊创始人贝佐斯大手笔投资人形机器人初创公司Figure AI，亚马逊旗下的Agility Robotics也在俄勒冈州建造了世界上第一个大规模生产人形机器人的工厂。

随着中国国内对机器人技术的重视和投入，人形机器人领域已涌现出了一批具有创新能力和市场竞争力的中国企业，中国人形机器人产业发展取得了显著的进步。3月北京人形机器人创新中心宣布近期将发布第一代通用开放人形机器人本体。

在九款亮相英伟达发布会的机器人当中，也有两款为中国公司得产品，分别是杭州宇树科技的H1和小鹏鹏行的PX5。

根据宇树给《环球时报》采访函的回复，H1是一款能跑的全尺寸通用人形机器人，配置了360°全景深度感知，目前移动速度达3.3m/s， 是全尺寸电驱动人形机器人的世界纪录保持者，其潜在移动速度达到5m/s。此外该机器人拥有更完善的全身动态协调能力，可以编组跳舞，甚至可以后空翻。因此，英伟达选择与宇树合作，共同推动全球AI机器人的深度进化。英伟达作为GPU和AI芯片领域的领军企业，则为宇树的机器人提供了强大的计算能力和深度学习技术支持。

但总体来说，AI与人形机器人的深度结合，在全球范围内都依然任重道远。

张锐认为硬件方面的挑战是一个重要的难点。尽管在算法层面，我们可以实现各种复杂的功能和性能，但在实际的机器人硬件上，却往往难以达到理想的输出功率和效率。这主要是因为硬件的设计和制造涉及到众多的技术细节和工程挑战，需要不断地进行优化和改进。

另一方面，目前AI 技术的进展主要局限于逻辑层面的深化与创新，而在思维与情感层面的突破尚显不足。机器人之形虽可塑，但其内在的思维与情感之“神”却仍是探索的疆界，要想与人类的情感达到70%的相似度，张锐认为还需要5-10年才可实现。

在3月初于斯坦福大学举行的经济论坛上，黄仁勋坛上预测能够通过人类测试的通用人工智能，或称“类人类”（artificial general intelligence，AGI，能表现人类所有的智慧行为）很可能在5年内就会出现。但黄仁勋也指出达成这一目标并不是没有难处，难处就在于科学家对于人类心智运作的方式仍然缺乏统一定义，“因此工程师会很难实现目标”。

4月8日马斯克在社交媒体上表示，到明年年底或者2026年，新的人工智能（AI）模型可能将超越人类的智力，“比最聪明的人还聪明”。去年他就预测，人类将在2029年“完全”实现通用人工智能。

有专家认为随着芯片和算法的不断进步，人工智能或许能最终超越人类智能。但北邮人机交互与认知工程实验室主任，人机混合智能专家刘伟指出通用人工智能或许是一个伪命题，并不是因为目前的人工智能系统还远未达到通用智能的水平，而是由于其本质上根本不能够像人类一样运用智能进行多种任务和学习。

通用智能的发展面临着技术性、生物性和社会性三大瓶颈。技术性瓶颈表现在人工智能系统需要具备更高的计算能力、更先进的算法和更有效的数据处理方法，以实现更复杂、更智能的功能。生物性瓶颈主要体现在我们对人类大脑认知能力运作机制的理解还非常有限，要实现类似的智能水平需要更深入的神经科学和认知研究。社会性瓶颈则包括了人工智能系统与人类社会的融合问题，例如文化差异、伦理道德、隐私保护等，这些都是影响人工智能发展的重要因素。

“克服这些瓶颈需要跨学科的合作和持续的创新努力，只有在技术、生物和社会方面取得突破，“通用智能”才能迈向更加成熟和全面的发展，但终究极难实现！或许根本实现不了！”刘伟说。

伦理安全风险

中国工业和信息化部2023年11月发布的《人形机器人创新发展指导意见》提出，计划到2025年，人形机器人创新体系初步建立，“大脑、小脑、肢体”等一批关键技术取得突破，确保核心部组件安全有效供给。整机产品达到国际先进水平，并实现批量生产，在特种、制造、民生服务等场景得到示范应用，探索形成有效的治理机制和手段；2027年，综合实力达到世界先进水平，成为重要的经济增长新引擎。

在技术和政策双方面助推下的人形机器人产业走上了发展的快车道，但同时也面临着安全，社会伦理和法律规范方面的挑战。

有专家指出，在一些需要密切接触的场景下，例如在养老或人员帮扶等需要肢体接触的场景中，人形机器人的安全性和风险问题不可小觑，还需要行业和有关监管部门在技术和应用场景上进行更多的研究工作。

张锐认为当人形机器人在家庭环境中出现意外状况时，由于目前缺乏明确的法律界定，使得责任归属和权益保障变得模糊不清。此外，关于人形机器人产品的合格性认证，当前尚未有强制性的制度予以规范，这无疑增加了产品批量发放的难度，使得人形机器人的广泛应用受到一定限制。因此必须正视并尽快解决这一法律层面的空白，以确保人形机器人在未来的发展中能够健康、有序地推进。

但他坚信，随着技术的不断进步和法规的逐步完善，人形机器人将在更多领域展现其独特魅力，为人类社会的发展贡献更多力量。

2023年，科技部会同教育部、工业和信息化部等10部门印发了《科技伦理审查办法（试行）》，其中提到，开展科技活动应坚持促进创新与防范风险相统一，客观评估和审慎对待不确定性和技术应用风险，遵循增进人类福祉、尊重生命权利、坚持公平公正、合理控制风险、保持公开透明的科技伦理原则，遵守我国宪法、法律法规和有关规定以及科技伦理规范。