Tencent Chief Scientist: Hierarchical Control Is Key to Embodied AI, Not Just LLMs

At the "AI-Human Relations Outlook" forum hosted by Tencent Research Institute and partners on October 10, 2026, Tencent chief scientist and director of the Robotics X Lab Zhang Zhengyou delivered a keynote on the challenges and progress of embodied AI.

Zhang argued that embodied intelligence requires a physical carrier to interact with the world, accumulating knowledge and skills through perception, control and autonomous learning. Simply putting large language models into robots will not produce genuine embodied intelligence, he said, stressing that intelligence and the body must be integrated.

He introduced the lab's hierarchical control approach, which separates control at the robot body, environment and task levels, allowing each layer to update knowledge independently. In 2026, the lab integrated its self-developed five-fingered dexterous hand, robotic arm and a mobile base into a robot for the first time. Combined with perception and planning foundation models, the robot can converse and complete tasks autonomously.

Zhang said robots will eventually enter households, and in the near term, areas such as rehabilitation and elderly care and personalized education may see the most immediate changes.