Yao Maoqing: World model training faces data mismatch and scale bottlenecks

robot
Abstract generation in progress
Mars Finance News: On July 17, Mifeng Technology CEO Yao Ouqing said at WAIC 2026 that the core of the world model lies in predicting the “next state” of the physical world. He admitted that current research largely remains at the level of visual generation, with two major pain points: first, much of the online video contains anti-physical content and lacks the contact-interaction data that robots need; second, the data scale is far inferior to that of large models, the information density in the physical world is low, and achieving high-level reasoning or judgment would require “100 million hours” of real-robot data. Yao Ouqing emphasized that a world model must have multimodal understanding, mastery of physical laws, and causal reasoning capabilities. When discussing deployment, he believes that in the short term, scenarios with high-frequency, rigid demand and environments that are controllable will see breakthroughs first; widespread adoption in open environments such as homes will depend on the industry building stronger generalization capabilities. (Wide-angle observation)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned