MoWorld 2.0 brings 4D world modeling toward industrial deployment
TMTPost reports that Magicore Technology and three national AI application pilot bases released MoWorld 2.0, a 4D world model built on Huawei Ascend hardware for embodied training, autonomous driving, industrial digital twins and other applications. Magicore says the model can represent more than ten physical factors, generate reusable 4D Gaussian assets and render them in real time, with company-reported inference speeds and deployment-cost reductions. The broader analysis says world models are moving toward production scenarios, but real-world data costs, real-time inference and the gap between simulated and physical behavior remain unresolved challenges.
MoWorld 2.0 illustrates the shift from generative content toward interactive, physics-aware simulation for embodied AI and industrial systems, although its performance and cost claims have not been independently verified.
(本文作者为 Leo张ToB杂谈,钛媒体经授权发布)
以具身智能为代表的物理AI是今年科技圈最火的一个词,目前也已经有不少的物理AI开始进厂了,但曾有某具身智能公司的CEO告诉我们,当前大多数进厂的具身智能,其实还不是具身智能,而只是在原本工业自动化设备上,加上一些AI的能力,真正的具身智能是能在不同产线、不同环境中,具备一定独立思考的能力,能自己感知世界的。”这点能力,当前绝大多数的机器人还不具备,所以还不能称之为具身智能。“该名CEO如是说。
无独有偶,在WAIC 2026上也有不少业内专家都曾提出,AI的核心命题正从生成,转向理解与交互的观点。而具身智能要具备理解和交互的能力,一个不可或缺的技术底座就是世界模型。
世界模型正在经历一个关键的阶段,往前跃升,既是真正解决“理解”世界和“模拟”世界的一跃,也是规模应用到具体生产生活场景的一跃。
#Magicore Technology#MoWorld 2.0#Huawei Ascend#Huawei Cloud#World Labs#Atlas#JEPA#Physis-v0.1#Pony.ai#Moore Threads#ω-EVA