Recently, Dr. Dong Xiaojian, Founder and CEO of Vizum, was invited to take part in XinhuaNet’s special interview series “Seeing 2026.” In the interview, he shared Vizum’s perspectives and practical approach to key questions facing the embodied intelligence sector—its progress in 2025, the most important breakthrough directions for 2026, and the essential conditions for humanoid robots to move toward scaled mass production and broad real-world adoption.

The Industry Enters an Acceleration Phase: From “Buzz” to “Validation and Deployment”

2025 is widely regarded across the industry as a milestone year for embodied intelligence. Driven by national strategic initiatives and strengthened cross-industry collaboration, the sector has achieved rapid advances and continued to reinforce its foundations for industrialization. Looking ahead to 2026, as core technologies continue to break through and commercialization deepens, embodied intelligence is entering a pivotal phase of accelerated evolution.

Scaling Up Production: Start with “Stability” and “Cost”

In Dr. Dong’s view, the prerequisite for “scaled mass production” is not simply building prototypes—it is ensuring robots can operate stably and sustainably in real-world environments. He emphasized in the interview that, for large-scale adoption, products must withstand the demands of long-duration operation. At the same time, cost optimization is a fundamental requirement for entering a wider range of scenarios. This calls for continuous iteration across processes, materials, and engineering systems, so that products can become viable for more service and application settings.

Behind this assessment lies a shared industry challenge: moving from “demo capability” to “delivery capability.” Stability determines whether a robot can keep working over time; cost and manufacturability determine whether it can be deployed at scale.

From Teleoperation to Autonomy: Humanoid Robots Need Human-Like Self-Perception

Dr. Dong further noted that true autonomy in humanoid robots requires stronger self-perception and self-diagnosis—the ability to sense internal states and detect issues, rather than relying on teleoperation to complete tasks. In other words, to reach the stage of large-scale application, key capabilities must be built into the system so robots can operate in complex, uncertain real-world environments with greater self-awareness, autonomous decision-making, and reliable execution.

“Operate First, Talk Later”: Stereo Vision as the Foundation of Manipulation

When discussing capability priorities for humanoid robots, Dr. Dong offered a clear engineering judgment: for humanoid robots, “manipulation” comes first. One of the critical enablers of manipulation is a high-precision stereo vision system that is interpretable and learnable. For real tasks such as picking, carrying, assembling, and organizing, vision must do more than detect—it must perceive clearly, understand the scene, and support closed-loop learning, enabling stable operation and continuous improvement.

As a company with long-term focus on stereo vision, Vizum has consistently invested in both R&D and engineering deployment of stereo vision systems. Guided by the goal of being “reliable in real operating conditions,” Vizum continues to strengthen an end-to-end capability framework—from core algorithms and integrated software-hardware systems to industry application solutions. By tightly coupling stereo vision with robot manipulation tasks, Vizum aims to ensure vision capabilities deliver measurable value in real workloads: more robust recognition and localization, more controllable motion guidance, and more sustainable iteration—providing essential support for the scalable deployment of humanoid robots.

From “impressive demos” to “reliable delivery,” and from “able to run” to “able to run reliably over the long term,” the embodied intelligence industry is entering a phase that places greater weight on engineering rigor and scaling capability. Vizum will continue to focus on stability, cost optimization, and the development of critical capability systems—advancing embodied intelligence toward validated, repeatable deployment paths across more real-world scenarios.