Embodied AI Revolution: Key Takeaways from HDC 2025 Roundtable

Physical Intelligence Takes Center Stage

In the wave of digital and intelligent transformation, physical intelligence has become a key driver of industrial innovation, spanning smart manufacturing to multi‑scenario robotics.

HDC 2025 Roundtable Overview

From June 20‑22, the Huawei Developer Conference (HDC 2025) in Dongguan hosted a roundtable titled “Connecting the Physical World, Co‑creating an Intelligent Future.” Over 120 scholars, researchers, and industry leaders attended.

Organized by Huawei Cloud and the Strategic Research Institute, participants included Ding Ning (Deputy Director of Guangdong Embodied Intelligence Robot Innovation Center), Luo Ziyan, Gao Jiyang (CEO of XinghaiTu), Prof. Guo Yulan (Sun Yat‑sen University), Prof. Li Meng (Shenzhen University), Huawei executives such as Ren Xudong, Cao Pei, Chen Xinyu, Lin Huading, Zhao Qiujing, Sun Gang, and others, with Zhao Weiming (Shanghai Jiao Tong University) as moderator.

Key Discussions

Embodied AI Challenges and Development – Ding Ning highlighted issues such as incomplete problem‑boundary definitions and uneven data distribution across scenarios, urging multi‑scenario data aggregation and leveraging China’s industrial strengths to advance general‑purpose robots.

CloudRobo Platform – Zhang Jing introduced Huawei Cloud’s CloudRobo platform, which integrates a multimodal digital‑twin data engine, agile training and validation tools, and an edge‑cloud collaborative robot‑cluster management system, enabling rapid development for industrial logistics.

Three‑Dimensional Scene Generation for Embodied AI – Guo Yulan presented three innovations: VideoDirector for controllable video generation, Layout2Scene for 3D semantic‑layout‑guided scene synthesis, and POSITION for efficient CAD model recombination, all aimed at providing controllable virtual environments for embodied agents.

Edge‑Cloud Collaborative Multimodal Quadruped Robots – Li Meng described a “cloud brain + edge cerebellum” architecture where a cloud‑based large model handles reasoning and planning, while lightweight edge models manage perception and execution, achieving superior navigation performance over existing methods.

Smart Agriculture Application – Ma Wenzhe from Southeast University showcased a drone‑plus‑tracked robot system for intelligent grape harvesting, combining multispectral imaging, BeiDou positioning, and multimodal data fusion to improve efficiency and resilience to extreme weather.

Conclusion

The roundtable demonstrated how academia and industry are jointly advancing embodied intelligence from theory to real‑world deployment, and Huawei pledged continued investment in research, ecosystem building, and talent cultivation to accelerate physical‑intelligence applications across sectors.