①Next year will be a critical year for the implementation of humanoid robots. Humanoid robots have been deployed in scenarios such as performances, interactions, and exhibition hall tours, but bottlenecks remain for entering factories to achieve 'large-scale delivery.' ②The robotics industry is moving from basic body manufacturing to more specialized and challenging fields, with future development extending further upstream towards greater refinement and higher technological intensity.
The STAR Market Daily (reporter Huang Xinqi) reported on November 8 that during the Eighth China International Import Expo (CIIE), experts from companies such as Unitree Robotics, Ubtech Robotics, and SenseTime shared their views on the practical application of embodied intelligence.
According to information obtained by The STAR Market Daily, next year will be a pivotal year for the deployment of humanoid robots. Currently, humanoid robots have already been applied in certain areas such as performances, interactions, and exhibition hall tours. However, bottlenecks persist for entering factories to achieve 'large-scale delivery,' and there remains a gap before true industrialization can be realized. If humanoid robots are to enter home environments in the future, they will need to address even more complex issues such as standard regulations, safety, and user privacy.
Overall, the robotics industry is transitioning from basic body manufacturing to deeper exploration in more specialized and challenging fields. Future development will continue to extend upstream, progressing towards greater refinement and higher technological intensity, ultimately building a complete humanoid robot industrial chain.
▍Next year will be a key year for the deployment of humanoid robots
Jiao Jichao, Vice President of Ubtech Robotics and Dean of its Research Institute, believes that next year will be a critical year for the deployment of humanoid robots. For the industry, especially for Ubtech Robotics, it is essential to identify a truly implementable scenario that is replicable and has a certain scale.
Regarding humanoid robots, Jiao Jichao believes that industrial scenarios should be the first to see deployment, enabling fully autonomous operations. This year, commercial scenarios have also seen genuine applications beyond dancing and performances, including guided tours, roles in 4S stores, and exhibition hall explanations.
"Next year, full-size humanoids will find broader applications in these scenarios. In other industries, it will take another 2-3 years for industrial scenarios to generalize into more complex operations. For commercial scenarios like hotel service staff roles, which involve more complex human interaction, this timeline should extend to 3-5 years. Household scenarios involve standardization and safety issues; humanoid robots capable of functioning in home environments will require at least 8-10 years."
Pan Zhengyi, Co-founder and Chief Operating Officer of Microbillion Intelligence, also believes that artificial intelligence will play a crucial, if not decisive, role in future smart manufacturing. The core challenge currently faced by manufacturing enterprises lies in: excessive reliance on automation may lead to rigid production, making it difficult to adapt to flexible demands involving multiple varieties and small batches; while over-reliance on manual labor can cause efficiency bottlenecks and poor quality consistency. The deep integration of embodied intelligence technology with industrial robots has become the key solution, such as through intelligent robot debugging and flexible switching between multiple processes, achieving a balance of production flexibility, efficiency, and quality, and truly driving the deep implementation and value realization of artificial intelligence in industrial settings.
Wang Xiaogang, Co-founder of SenseTime, stated that the current large-scale deployment of humanoid robots primarily focuses on providing emotional value, such as performances, interactions, and exhibition hall explanations. Over the next three years, standardized scenarios like industrial settings and logistics sorting are expected to achieve initial breakthroughs.
Moreover, with the development of autonomous driving technology, unmanned logistics vehicles are ushering in an opportunity for rapid adoption, which will address the 'last mile' end-stage delivery challenge in depth. They can independently complete tasks such as loading goods and sorting, and be applied to scenarios with high standardization and large scale, such as forward warehouses and flash purchase warehouses. However, their widespread application still faces a core challenge: how to accurately identify and process thousands or even tens of thousands of SKUs (stock-keeping units), which poses extremely high demands on the visual perception and cognitive decision-making capabilities of machines.
Xu Jincheng, founder and CEO of Pasini Perception Technology, is optimistic about the implementation of near-end small humanoid robots. "Small humanoid robots currently already have substantial implementation capabilities in entertainment-related and education-related fields. Entering factories may happen within five years."
Li Tong, founder and CEO of Keenon, highlighted the overseas opportunities for Chinese robotics. "At this critical juncture, robots are entering factories and daily life globally on a large scale, providing an unprecedented overseas window period for Chinese robotics companies."
Li Tong pointed out that, in the AI era, China's accumulated manufacturing and scenario advantages are transforming into overseas expansion advantages. "Robots, as carriers of hardware AI, whether in manufacturing, logistics, healthcare, or service industries, allow Chinese robotics companies to provide internationally competitive solutions."
▍Mass delivery of humanoid robots remains challenging
Li Zhiqiang, founder and CEO of Yimoo Technology, stated in an interview with The Science and Technology Innovation Board Daily that the core of embodied intelligence lies in being 'human-like,' which requires a complete closed-loop control system encompassing perception, decision-making, and execution to achieve human-like embodied intelligence. Among these, perception ability is essential and highly important. Yimoo Technology previously launched the world’s thinnest commercially available bionic visuo-tactile sensor, and has already garnered sales and cooperation leads from hundreds of customers.
He believes that Chinese enterprises have significant opportunities in emerging fields like visuo-tactile sensors. Nevertheless, improvements in parameter performance and mass production capabilities are still needed, as companies capable of achieving mass production at levels of 100,000 or even millions remain relatively scarce.
Discussing industry trends, Li Zhiqiang noted that the robotics industry is exploring finer and more challenging directions. "This year's trend shows increasing degrees of freedom and flexibility in robotic hands. In the future, the industry will continue to move upstream towards more refined and technologically advanced directions, ultimately forming a complete humanoid robot industrial chain."
In Li Zhiqiang’s view, bottlenecks still exist in the mass delivery of humanoid robots at this stage, and true industrialization remains some distance away.
"Industrial applications involve less complex movements but require high precision in repetitive operations, where traditional robotic arms can often suffice. Such scenarios are not the most suitable for humanoid robots. Yimoo Technology is exploring non-repetitive, non-standardized scenarios requiring precise operations."
Regarding the specific path for the implementation of embodied intelligence technology, Pan Zhengyi emphasized that at the hardware level, the technology for industrial robot bodies is already relatively mature. The current core lies in integrating upstream and downstream ecosystem resources, such as collaborating with manufacturers in areas like end-effectors, to enable flexible task switching and efficient adaptation of robots across different application scenarios. At the software algorithm level, whether it involves perception, motion, or final execution algorithms, the key is to use algorithmic data to consolidate, replicate, and transform human-accumulated craftsmanship experience, propelling robots from being 'execution tools' to becoming 'intelligent partners' capable of autonomous learning and flexible adaptation. This is also the key distinction between true artificial intelligence and traditional automation technologies.
Leng Xiaokun, Chairman of Aelos Robotics, predicted that next year humanoid robots in the industrial sector could see companies with procurement and delivery volumes exceeding ten thousand units. It is expected that within the next five years or so, related technologies will likely surpass the 'basically usable' threshold.
"However, for them to truly enter our daily lives as products, it may take a decade-long process. Issues involving standards, safety, privacy, and more are far more complex and challenging than technological problems."
▍World models are viewed more favorably
Currently, world models and VLA models represent two parallel and complementary technical approaches in the field of embodied intelligence. Li Zhiqiang, founder and CEO of Yim Technology, believes that it remains undecided which approach will ultimately prevail. "In the medium term, VLA models serve as supplements; in the long term, there is a high probability of convergence towards world models, although this depends on extensive data training."
Wang Xingxing, founder of Unitree Robotics, also favors world models based on video generation. He pointed out that current models based on VLA+RL still fall short in terms of generalization capabilities, and he personally prefers world models constructed using video generation technology.
However, he also noted that world models face challenges. "Small and medium-sized robotics companies struggle to run these models because video generation models require significant computational power and more powerful computing cards. Currently, large AI and internet companies possess richer resources for video models, making it more likely for them to successfully develop such models."
