Humanoid robotics is undergoing a paradigm shift, moving from performance-driven showpieces toward practical, scalable industrial applications and standardized consumer products, according to industry leaders speaking at the Boao Forum for Asia Annual Conference 2026 on Wednesday.

At a sub-forum focused on the future of embodied intelligence, executives and scientists said while significant challenges remain — particularly in data acquisition, safety and core "brain" functions — China's unique industrial ecosystem positions it to lead the next phase of the industry.

Xiong Youjun, CEO of the Beijing Humanoid Robot Innovation Center, detailed the technical evolution dividing humanoid robotics into three dimensions, namely: the physical body meaning hardware; the cerebellum for movement control; and the brain for cognition and decisionmaking. While hardware architectures are beginning to converge, the most significant progress is in the synergy between the cerebellum and the brain, driven by advances in large language and multi-modal models.

"The co-evolution of the cerebellum and brain is very evident," Xiong said. He pointed to his center's achievements in movement control — such as a robot autonomously completing a half-marathon in Beijing last year — and in intelligent manipulation through proprietary models. Crucially, Xiong emphasized that these technologies are being opened to the broader industry, allowing universities and developers to accelerate collective progress.

A key theme across presentations was the rejection of what speakers called "performance-style showboating" in favor of real-world utility. Xiong outlined how the industry is pivoting toward applications that serve "the main battleground of the national economy". Early adoption is occurring in structured environments like automotive and home appliance manufacturing, logistics and sorting.

Shao Hao, head of the robot lab and chief scientist at Vivo, added a cautionary note on data — the critical bottleneck for achieving a "ChatGPT moment" in robotics. Unlike the one-dimensional text data that fueled large language models, robotics requires high-dimensional data, creating a monumental challenge for scalable training. He argued that the solution lies not in costly bespoke data collection, but in tapping into vast, existing sources like human video data.

Shao also focused on the safety and regulatory challenges of bringing robots into homes — Vivo's ultimate target market.

Addressing skepticism about humanoid forms, Shao argued that "application scenarios define forms". While specialized machines like robotic vacuums excel at single tasks, the humanoid form is uniquely suited to environments built by and for humans. Over the long term — a decade or more — it will outperform specialized robots in versatility.

Shen Dou, executive vice-president of Baidu and president of Baidu AI Cloud Group, identified three ongoing bottlenecks: the physical body, data and models. However, he expressed strong confidence that China's complete supply chain, vast pool of engineers and world-leading cost-control capabilities will transform embodied intelligence into a standardized industrial product category, significantly lowering costs.

"The future cost of robots will drop significantly," Shen said, adding that pricing models may even evolve toward token-based consumption, similar to cloud-based AI services.

While acknowledging that industrial scenarios remain easier to standardize than home environments, he projected that costs will eventually fall to household-consumer levels.