X Square Robot builds a foundational stack for general-purpose robots

X Square Robot, a Chinese embodied-AI company, argues that the recipe for general-purpose robots is an integrated stack spanning data, a world model for predicting physical changes, and an action model that unifies perception, planning, reasoning, and decision-making. The stack is built on three principles: the basic unit of robot data is an interaction, not a trajectory; pretraining should yield usable capability; and behavior should be modeled around physical events rather than fixed time slices. For data collection, the company uses its Universal Manipulation Interface (UMI) system, QUANXTA Zero Series, which captures demonstrations via a wearable rig with dual grippers. It emphasizes quality control by physically replaying trajectories and counting only those that complete tasks as valid. The world model, WALL-WM, treats action-grounded semantic events (e.g., reaching, grasping) as units, coupling a text-to-video model with a fresh action network to preserve visual priors. The action model, Wall-OSS-0.5, is a vision-language-action model that must run on a real robot before task-specific fine-tuning, and uses X-Tokenizer, a semantic interface for discrete action tokens aligned with language model features. X Square Robot's valuation has climbed above 20 billion yuan (about $2.9 billion).