Beijing-based Galbot delivered one of the most talked-about demonstrations at WAIC 2026 in Shanghai, where crowds three to four layers deep gathered to watch its G1 humanoid robot prepare breakfast with remarkable adaptability.
A Breakfast Demo That Went Beyond the Script
The Galbot G1 performed routine breakfast tasks — toasting bread, pouring drinks, and arranging tableware — with the smooth coordination of an experienced short-order cook. But what truly captivated the audience was not the actions themselves; it was what happened when spectators actively interfered.
Attendees repeatedly moved the bread aside, shifted the kettle, rearranged cups and plates, and even covered cup lids with their hands, creating obstacle after obstacle. Yet the robot never wavered:
- Millisecond-scale target relocation — re-identifying moved objects in real time
- On-the-fly path reconstruction — recalculating motion trajectories around new obstacles
- Zero interruption completion — finishing every step despite continuous interference
This level of dynamic adaptability represents a meaningful step forward for humanoid robots, which have historically struggled with unstructured and unpredictable environments — a key barrier to real-world deployment.
From Conference Demo to Policy Attention
The significance of Galbot's technology extends well beyond the exhibition floor. The company was among the invited enterprises at a private enterprise symposium hosted by Zheng Shanjie, head of the National Development and Reform Commission (NDRC), focused on analyzing first-half economic performance and gathering suggestions for second-half economic work.
Wang He, founder and CTO of Galbot, outlined a roadmap suggesting that the humanoid robot industry could reach its "ChatGPT moment" by late 2028, based on current data accumulation rates and model convergence trends. He emphasized that a powerful pre-trained model combined with an efficient post-training paradigm is the key to unlocking this milestone.
The "Brain and Cerebellum" Architecture
Wang described building a unified "brain and cerebellum" — an embodied intelligence foundation model that achieves a deep understanding of human movement and control mechanisms — as a critical technological pathway for pushing humanoid robots toward the tipping point of artificial general intelligence.
Industry context underscores the urgency: China's humanoid robot output reached about 20,000 units last year, surpassed 40,000 units in the first half of 2026, and is expected to exceed 100,000 units this year — a fivefold jump signaling acceleration from technological exploration toward standardized, large-scale development.
As the industry moves from single-unit prototypes and stage demonstrations toward batch deliveries, the coordination requirements across research, production, testing, application, and safety governance rise dramatically. Galbot's WAIC demonstration suggests that at least one piece of the puzzle — real-time adaptability in unstructured environments — is progressing faster than expected.
