Instead of combining separately trained modules, the World Unified Model trains a single network that jointly represents: perception (processing sensory data), language (instruction understanding and reasoning), action (robot motion control) and physical prediction (anticipating the effects of actions in the world). Through joint training, the model builds a coherent representation in which reasoning and acting are grounded in the same space. The architecture is realized in the WALL models (e.g. WALL-B) at a billion-parameter scale, supporting generalization across tasks and scenarios.
Classical robotic systems combine separate vision, language and control modules, which makes coherent information flow between them and generalization difficult. WUM integrates these functions into a single network, enabling natively multimodal, end-to-end control and incorporating physical prediction (the consequences of actions) within one model.
Processing sensory data (vision and other modalities) within the shared network.
Understanding natural-language instructions and reasoning (including spatial and causal).
Generating precise robot control actions, including force control and positioning.
Predicting the effects of actions in the world (a world-modeling element) trained in the same network.
A single jointly-trained network replacing separate vision/language/action modules.
A unified end-to-end network is harder to analyze and fix than a modular architecture with separated components.
Joint multimodal training requires large, diverse embodied data and significant compute.
Many results and claims come from the vendor and require independent verification.
X Square Robot’s open-source vision-language-action model (September 2025) — an earlier stage of the WALL series.
The WALL-B model (April 2026, Beijing) introduces the World Unified Model architecture, training perception, language, action and physical prediction in one network.
Jointly training perception, language, action and physical prediction in one billion-parameter network is compute-intensive and requires large, diverse embodied data.
Model size; WALL models operate at a billion-parameter scale.
Which functions are trained jointly in one network.
A unified, dense end-to-end network processes all modalities in a single pass, without separate modules.
Training the large network is highly parallel on GPU clusters, but the perception–action control loop on the robot is inherently sequential in time.
Training and inference of a large unified embodied model require GPU acceleration.
