WAIC 2026 Roundtable: Generalized Embodied Intelligence Must First Break Through Specific Scenarios, with Future Competitive Focus Shifting to High-Quality Data Acquisition and Scene-Centric Validation

According to Dongcha Beating monitoring, Fudan University Vice President Jiang Yugang, Wisdom Element Robot Partner Yao Maoqing, EtStone Intelligent Navigation CEO Chen Yilun, and BrightOrigin CEO Jiang Xu participated in a roundtable discussion at the 2026 World Artificial Intelligence Conference, focusing on the world model. The panelists agreed that the essence of the world model lies in understanding the laws of the physical world to predict the next state or action, rather than merely rendering images. It requires native integration of multimodal fusion, physical laws, causal reasoning, and long-range prediction capabilities. The current major bottleneck lies in data—Chen Yilun pointed out that video data lacks key modalities such as force and touch. Ideal training data needs to meet three conditions: multimodal completeness, high-frequency interaction, and real-world origin, with embodied intelligence requiring complex operation or millions of hours of real interaction data. Yao Maoqing drew a parallel to the billion-hour speech training volume of large language models and estimated that the physical world may require "over a hundred million hours" of real data to master common-sense physical prediction. At the architectural level, Jiang Xu pointed out that the current mainstream architecture conflates state prediction with action prediction, leading to a conflict between generative and understanding abilities, making simultaneous optimization difficult.

In terms of implementation pathways, all three guests view the manufacturing industry as the most predictable scaled scenario for the next three years:

Yao Maoqing revealed that Wisdom Element Robot has achieved robot swarm operations on the production line with six days, sixty thousand operations, and a 99.99% success rate;

Chen Yilun is betting on the manufacturing industry, citing reasons such as high data density, clearly defined task completion standards, and abundant human demonstration data. EtStone Intelligent Navigation has collaborated with automakers to advance the deployment of a thousand-unit industrial embodied robot cluster, emphasizing that China's manufacturing sector is the most concentrated globally, making it an ideal testing ground for physical AI;

Jiang Xu believes embodied intelligence is an extension of multimodal large models. With the internet already having 100 billion hours of video data suitable for pre-training, a breakthrough in capabilities will first appear in daily scenarios such as homes and offices. However, commercialization needs to meet high fault tolerance conditions, and finding scenarios for large models is no easier than training models.

The consensus among the three is that we are still far from achieving general embodied intelligence, and breakthroughs in specialized scenarios are a necessary stage. The future competitive focus will shift from model architecture to high-quality data acquisition and closed-loop scenario validation capabilities.

---------------------------------

Click the original text link below to join Dongcha Beating · Feishu AI News Channel, which continuously monitors global AI hotspots and news 24/7.