SHANGHAI — With pitch-black mechanical legs and an expressionless face save for somewhat animated eyes, the 5.5-foot, 75-kilogram robot XR4 could have been plucked straight from a sci-fi movie, albeit one about domesticity. Like humans, it can understand conversations, pick up an egg, and even bake bread.

Developed by the Chinese startup Dataa Robotics, the XR4 was among the standouts at the World Artificial Intelligence Conference (WAIC) that kicked off in Shanghai on Thursday.

Described as a “breakout” event by business insiders, this year’s World Artificial Intelligence Conference showcased robots of varying intelligence and agility, some already serving as helpers in schools and hospitals. Now in its seventh year, the WAIC brings together executives, scientists, and scholars to discuss the most pressing issues in AI, with a special focus on safety and governance.

Some of the most advanced humanoid robots, including Fourier’s GR-1, now produced at scale for commercial use, and Leju’s Kuavo, powered by Chinese tech giant Huawei’s Pangu large language model (LLM) and applicable to scientific research, education, and logistics, are at the event. Tesla also presented its second-generation Optimus, claiming the upgraded model can master tasks as precise as boiling an egg.

While robots have existed for decades, humanoid robots, mimicking human appearance and behavior and paired with rapid advancements in AI, are a recent development. The technology has reached new heights with the creation of “soft” human-like skins.

Humanoid robots are expected to take on roles in industrial production and daily life, particularly in repetitive, unengaging, and dangerous tasks.

Yet, mastering humanoid robotics presents unprecedented difficulties compared to other types of robotics. Just keeping a two-foot structure balanced while walking like a human has been an arduous technical challenge.

“A full-size humanoid is even more difficult to maneuver, but it holds more value for its seamless integration into most life scenarios,” said Xu Bin, general manager of the National and Local Co-Built Humanoid Robotics Innovation Center, a research institute in Shanghai.

But widespread use may be closer than we think. “In three years, the technology could achieve large-scale commercial applications,” Chen Yuan, vice president of Dataa Robotics, told Sixth Tone.

Growth and gaps

A race to master humanoid robotics technology is raging across the globe, with venture funding for startups in the field reaching a record high this year. In February, Goldman Sachs estimated the market could be worth $38 billion by 2035, a six-fold increase from the projected $6 billion a year ago.

Despite a late start, China aims to become a powerhouse in humanoid robots, integrating them into every aspect of society. Last year, authorities established a two-phase national plan to develop the field.

In November, the Ministry of Industry and Information Technology issued guidelines aiming for a “safe and effective” supply of key robotics components by 2025 and a robust industrial and supply chain by 2027.

That same month, Beijing set up the nation’s first provincial humanoid robotics innovation center to tackle major technical challenges, followed by the National and Local Co-Built Humanoid Robotics Innovation Center in Shanghai in May.

According to Xu, the Humanoid Robotics Innovation Center recently introduced Healthy Loong, a full-size humanoid robot prototype that will be open-sourced to all developers. This year, the center aims to establish a training base for robotics, where, by 2027, around 1,000 robots can be trained simultaneously.

“The goal is to further popularize the technology among enterprises to reduce development costs,” said Xu.

By February, there were around 25,000 companies involved in humanoid robotics in China, according to Chanyeos, an industrial information service platform. Last year, China surpassed Japan in filing the most patents related to this technology.

According to Chen from Dataa Robotics, Tesla’s innovations and achievements with its Optimus robots have sparked noticeable interest among Chinese companies.

The second-generation Optimus, which debuted at the WAIC, was first released in August 2021. Equipped with Tesla’s neural network and computer vision technology, it is designed for tasks like carrying heavy objects or even grocery shopping.

The past year has seen significant breakthroughs from homegrown pioneers too in the commercial applications of humanoid robots. For instance, the Beijing-based startup UBTech announced recently that its Walker S robots had entered the factories of major carmakers like Nio for training, demonstrating progress comparable to Tesla.

Another startup, Unitree, shocked the industry in May by selling its Unitree G1 robot, which can simulate human hands for precise operations, at an unprecedentedly low price of 99,000 yuan ($13,600).

Despite recent advancements, industry experts caution that there’s still a long way to go.

Xu underscored that the high cost of humanoid robots poses a significant challenge. Most robots are priced between 500,000 yuan and 1 million yuan, but Xu suggested that applications would surge if developers could reduce the price to around 200,000 yuan.

“The industry is yet to achieve large-scale production,” explained Xu. “Also, the current technology related to LLMs is not advanced enough, which sets high requirements for hardware. As a result, manufacturing is complex and costs are elevated.”

Despite strong demand for labor and automation, the market for robotics products remains relatively weak. This sales challenge was highlighted when Tesla revealed it had only manufactured five to six Optimus units over the past few years. Similarly, UBTech sold just 10 units of its Walker humanoid robots out of 760,000 total robots before the company went public last year.

Enterprises need to collaborate with suppliers and stakeholders to create products tailored to specific applications, a process known as “secondary development,” said an executive at a leading robotics firm, requesting anonymity since they were not authorized to speak to the media. “Many clients purchase robots for testing or research rather than immediate use in warehouses and factories,” the executive added.

For the industry to progress, enterprises must focus on innovating key technologies, Guo Qianqian, chief machinery analyst at SDIC Securities, told Sixth Tone. “Though LLMs have enhanced robotic intelligence, they still struggle with recognizing messages in various forms and making real-time responses,” said Guo.

Xu emphasized the importance of developing a mature ecosystem to achieve these ambitions, saying: “The chain from demand to industrial implementation to application deployment for humanoid robots has not been fully connected, so it hasn’t formed a positive feedback loop.”

Editor: Apurva.

(Header image: Dataa Robotics machines dancing at the 2024 WAIC in Shanghai, July 4, 2024. VCG)