Digital Huaxia's Chen Junmin: After Legs and Hands, Humanoid Robots Need Brains
Humanoid robots have become more capable over the past two years, with walking, running, dancing and grasping quickly becoming standard. As each action turns into an industry baseline, a new question arises: can robots enter hospitals, banks and libraries and complete tasks reliably over time?
Chen Junmin, cofounder of Digital Huaxia, sums up the rhythm: 2024 was about training legs, 2025 about hands, and by 2026, with hands and feet somewhat capable, it is time to train the brain. He says the next stage is not about showing off motion but about entering real scenarios and getting things done.
Different forms suit different scenarios, and bipedalism is not the only answer. Digital Huaxia is developing bionic, general-purpose and dual-form robots. Xialan, its bionic robot, targets retail, reception and interaction; Xiaqi, its general-purpose robot, combines interaction with some operational ability; Xingxingxia, its dual-form robot, supports both bipedal and wheeled movement. Chen says the layout is based on scenario differences.
In public spaces such as hospitals and libraries that need long periods of movement and unattended operation, relying solely on bipedal robots still poses challenges in stability, battery life and safety; wheeled movement is often more efficient. The company keeps bipedal motion control while adding wheeled forms, balancing technical advancement and practical application. Different bodies correspond to different scenarios, while some underlying capabilities are managed by a unified platform to reduce duplicate development. Its bionic robot has reached a third generation, with better multilingual interaction and fully managed services, and is moving from labs and exhibitions into commercial services.
A successful demo is not hard; stable daily work is. Moving from prototype to mass production crosses a threshold completely different from the demo stage. Chen cites the bionic head: one expression involves complex timing among multiple servos. Rather than presetting a smile, the company wants the robot to judge a person’s state from cameras, voice and other multimodal information, then drive expressions through an emotion model. In real scenarios, stability requirements rise further.
Cost is another issue. Prototypes can validate technology regardless of cost, but commercial products must bring costs down. Chen says Digital Huaxia positions itself as a service robot with warmth. Humanlike appearance and expressions are means to natural interaction, not the goal. It will not simply pursue an appearance infinitely close to a real person, focusing more on capabilities directly related to interaction.
Chen says some model capabilities may be overestimated, while engineering and data have long been underestimated. Many papers or demo videos make robots seem highly autonomous, but in real applications costs remain high and generalization is weaker than shown. Entering the market requires solving reliability, stability, cost, manufacturing and long-term operation. These abilities attract less attention than a new model but determine whether a product can land. Real physical-scenario data is also underestimated. Large models have relied on internet data, but embodied intelligence needs large amounts of real-environment data. Homes are open and complex; relatively closed settings such as banks and government service halls may be better for accumulating real-world physical data now.
Digital Huaxia’s products have entered banking, government affairs, healthcare and elderly care, but Chen admits that if full ROI is the standard, the entire industry has not truly solved the problem. Robots have mostly been used for performances and displays; on the surface they perform many actions, but staff often provide continuous support behind the scenes. Counting labor and operations, return on investment is not easy to establish. The company has explored public-service scenarios extensively and joined government-backed challenge programs, but overall it remains in a polishing stage.
In business models, hardware sales should remain the main approach short term. Digital Huaxia is exploring long-term leasing and service models. For government affairs and education, three-year or five-year cooperation periods are possible; customers receive not only hardware but also software upgrades, scenario capabilities, technical services and maintenance. Chen says commercialization ultimately must answer one practical question: can the value created cover the cost? The whole industry is still exploring.