生成预测四大权威榜单第一！开悟世界模型确立具身智能世界模型新标杆
Quick Answer
Kairos, developed by ACE Robotics, tops global benchmarks in embodied intelligence, achieving first place in RoboTwin 2.0 and others, showcasing a revolutionary unified architecture for video generation and state prediction.
Quick Take
This model, leveraging over 100,000 hours of human-centric data, enables real-time robotic responses with a 96.1% success rate in complex tasks.
Key Points
- Kairos achieved first place in RoboTwin 2.0 with a 96.1% success rate.
- The model integrates understanding, generation, and prediction in a unified architecture.
- It utilizes over 100,000 hours of human-centric real-world data.
- Kairos-4B is the first embodied world model for direct robot control.
- The architecture eliminates translation delays, enhancing real-time performance.
Source Excerpt
近日，大晓机器人开悟世界模型（Kairos）同时在 RoboTwin 2. 0、LIBERO-Plus、WorldModelBench Robot、DreamGen等全球针对世界模型视频生成和状态预测的权威具身智能评测中均实现第一，超越 Cosmos3、PI、MotuBrain、being-H0. 7、Abot、Fast-WAM、Wan2. 2等全球主流世界模型 。 凭借首创的原生统一世界模型架构——“多模态理解—生成—预测”一体化架构，实现了从具身视频生成、任务指令追随、泛化能力到场景鲁棒性的全维度领先，并向全行业开源。 针对行业普遍基于视频生成模型做后训练所固有的物理缺失、因果匮乏与推理延迟等痛点，大晓机器人（ACE ROBOTICS）于2025年12月重磅推出创新的一体化统一架构。 作为国内唯一坚持自主设计网络架构与自主预训练范式的团队，大晓彻底摒弃了“外挂式”改良捷径，从底层重构世界模型运行逻辑。 该架构深度融合理解、生成与预测三大核心能力，依托自研混合线性注意力算子与全局状态共享机制实现全链路高效协同。 这一颠覆性设计确立了全球主流演进方向，英伟达最新发布的Cosmos 3.
0亦采用同源架构，印证了Kairos的绝对引领地位。 在数据层面，得益于完全自主的预训练范式，Kairos广泛汇聚十余万小时human-centric实景数据与数百万小时互联网真实世界视频，全面覆盖数百类职业场景。 结合显式模仿学习与隐空间强化学习，模型彻底打通了从数据感知到深度理解的闭环，使基模型的场景认知、时序因果推演与泛化适配取得全方位突破。 …
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from 雷峰网 AI
See more →
刚刚，GPT 5.6 发布会上，OpenAI 暴露了哪些 Agent 技术路线？
OpenAI's GPT 5.6 integrates ChatGPT and Codex, introducing a for complex task execution, with models Soul, Terra, and Luna for efficient workflow management. The release emphasizes task orchestration, contextual understanding, and robust security measures for enterprise applications.
