AgiBot Launches GE-Sim 2.0 to Advance Embodied Intelligence and World Models

China’s AgiBot is focusing on embodied intelligence and seeking to use World Models to create digital environments capable of understanding space, movement and physical causality. The approach is intended to address the limitations of large language models, or LLMs, which learn primarily from text and struggle to grasp real-world physical laws. Such technology could serve as the digital brain of “physical AI” systems such as autonomous robots, making it critical to continuous training and autonomous decision-making in humanoid robots.

As of July 20, 2026, AgiBot had released its GE-Sim 2.0 world simulator, which uses generative simulated worlds to accelerate robot learning, testing and self-evolution while advancing embodied intelligence and World Model development. However, training and inference for such systems remain heavily dependent on advanced AI chips. Computing-power availability, costs and access to hardware will be key bottlenecks to deployment at scale.