Daxiao Robotics and Nanyang Technological University S-Lab Release World Model Mundo

Recently, Daxiao Robotics, together with Nanyang Technological University's S-Lab, released Mundo, a world model built for specific environments that can be explored and interacted with. The model supports scene reconstruction, free-viewpoint exploration and spatiotemporal simulation capabilities similar to Atlas, and incorporates physical properties such as object mass and friction into scene programs, computing motion, contact and collisions through a physics engine; it also supports invoking agents to edit objects, properties and actions in the scene through code, turning a small amount of imagery into a three-dimensional world that can be freely explored, programmatically interacted with, and used for robot navigation and manipulation training. Mundo combines three-dimensional scene representation, physical simulation and video generation, and together with Astra creates a new spatial understanding capability: it first reconstructs scenes and objects from imagery, then edits the scene and defines object properties and actions through code, has the physics engine compute motion and collisions, and has the video model render results in combination with reference imagery, forming a complete pipeline from scene reconstruction and code editing to interactive simulation. The scene can both be seen and respond to actions and record changes, becoming a training and testing environment for robots. In Daxiao Robotics' world model technology system centered on Kairos, Mundo supplements environment modeling, training data generation and simulation training capabilities, helping robots complete scene adaptation and policy optimization. In scene construction, Mundo can expand a small number of images or videos into a three-dimensional space that can be continuously explored. The same set of reference images can support multiple camera paths; for larger scenes, it can dynamically select context and continuously generate tens of seconds to minute-level exploration footage. Starting from contextual imagery of real places, it first builds a three-dimensional proxy, determines the spatial relationships between different views, and then generates new viewpoints along a specified camera path. In the demonstration scene, inputting 1 to 12 images allows exploration around the room's furniture and spatial structure. During modeling, a three-dimensional detector provides initial positions and extents for major objects, and a vision-language model agent further adjusts the proxy scene, through rendering, comparison and correction, making the scene structure correspond to the input imagery. When the camera moves, the input images constrain the already-captured layout, appearance and details, and the model infers and completes areas that were not captured. This capability can also start from a single generated image or painting, extending structure and visual style to new viewpoints. For larger environments, Mundo uses dozens of contextual images, generating in segments along the same three-dimensional proxy, selecting relevant reference views for each segment, and connecting successive segments through overlapping frames. In physical interaction, Mundo can edit the world through code, describing object names, sizes and poses through scene programs, and configuring physical properties such as mass and friction. Scripts define actions and applied forces, and the physics engine computes the objects' responses under gravity and contact constraints, obtaining motion trajectories, contact positions and collision forces; then, combined with environment reference images, the changes are rendered into the corresponding scene. Users can push and pull objects, open drawers, and also change materials and gravity conditions to observe responses. States such as velocity, contact points and impact forces can be displayed in sync with the generated imagery. The three-dimensional proxy and scene programs record object positions and changes, the physics engine computes motion, and the video model presents visual results, jointly forming an interactive environment. In robot training and testing, Mundo can support robot movement and manipulation within the same customized scene. For navigation, the three-dimensional proxy provides information such as object categories, positions, obstacles and traversable areas, and a path planner finds routes accordingly; Mundo can generate first-person and third-person footage of different robots traveling along the route. For manipulation, Mundo establishes geometric and physical properties for interactive objects, allowing robots of different forms to attempt the same task, such as putting an apple into a bowl, placing a can onto a tray or stacking blocks, and generates external views as well as robot head and wrist camera views, preserving task trajectories and physical interaction states. Visual observations and interaction states correspond to each other, making it easy to analyze the execution process by combining what the robot "sees" and what it "does", and to compare and adjust policies before entering real scenes. Mundo also extends spatial exploration into the temporal dimension: for dynamic events recorded by multiple cameras, it can present them from new viewpoints; for simulated events built from images, it computes motion through gravity and contact dynamics; by pausing time at a selected moment and moving the camera, the same instant can be observed from multiple angles, creating a "bullet time" effect that helps inspect action details and occlusion relationships. In a hotel application, the team has used Mundo to reconstruct floors and laundry areas, providing spatial and semantic information for navigation, and to simulate interactions such as opening a washing machine door. The industrial value of a world model must be demonstrated by its contribution to real tasks. In Daxiao Robotics' world model technology system centered on Kairos, Mundo supplements specific environment modeling and simulation training capabilities, connecting on-site data collection, data generation and policy optimization, and serving robots as they enter actual work scenarios.