Daxiao Robotics and Nanyang Technological University S-Lab Release Mundo, a World Model for Embodied Intelligence
大晓机器人联合南洋理工发布Mundo，具身智能世界模型核心能力实现突破
The model supports full indoor space modeling from one to twelve images and dynamically matches context for large scenes. It incorporates physical attributes like mass and friction coefficients into its programmable system, simulating interactions such as pushing, colliding, and falling. By generating continuous exploration footage and retaining task trajectories with physical interaction states, Mundo provides low-cost, high-fidelity virtual training data for reinforcement learning, addressing industry pain points where traditional visual generation models cannot respond to physical actions.
Source: Chinese robotics — Company news discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
image count for scene reconstruction
1–12
View original evidence
从核心技术能力来看，Mundo仅需少量图片或视频素材，即可快速重建可自由探索的三维场景，支持任意视角切换与长路径画面生成，输入1至12张图像就能完成完整室内空间的建模，面对大场景可动态匹配上下文素材，连续生成数十秒至分钟级的探索画面，大幅降低机器人场景适配的建模成本。
Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
从核心技术能力来看，Mundo仅需少量图片或视频素材，即可快速重建可自由探索的三维场景，支持任意视角切换与长路径画面生成，输入1至12张图像就能完成完整室内空间的建模，面对大场景可动态匹配上下文素材，连续生成数十秒至分钟级的探索画面，大幅降低机器人场景适配的建模成本。
Open source S4
区别于传统场景重建方案，Mundo首次将物体质量、摩擦系数等物理属性纳入场景可编程体系，通过内置 物理引擎 精准模拟推拉、碰撞、坠落等真实物理交互过程，所有运动轨迹、接触点、冲击力等状态数据均可完整记录，实现“所见即所动”的可交互仿真环境，解决了传统视觉生成模型无法响应物理动作的行业痛点。
Open source S5
在机器人训练场景中，Mundo可在同一定制场景内贯通导航、操作全流程，既支持路径规划生成不同视角的机器人移动画面，也能为不同形态的机器人分配物体交互任务，完整保留任务轨迹与物理交互状态，直接输出可用于 强化学习 的训练数据，大幅降低真实场景下的机器人试错成本。
Open source S6
Kairos是大晓机器人自主研发的具身智能世界模型体系，Mundo作为其环境建模与仿真训练模块的补充，补齐了从场景重建到训练数据生成的关键能力，帮助机器人快速完成陌生场景适配和运动策略优化，为后续具身智能机器人规模化落地工业、家庭等商业场景提供核心技术支撑。
Open source S7
问AI · Mundo作为补充模块怎样优化机器人运动策略？
Open source S2
Source：Chinese robotics — Company news discovery · news.qq.com