Inside WALL-A, a proprietary foundation model

Where many robotics teams fine-tune open-source language or vision models, X Square Robot is taking a more self-reliant route. Its core system, called “WALL-A,” is a proprietary, end-to-end foundation model built on a Transformer architecture that ingests visual, linguistic, and tactile signals at once.

In practice, that means the model doesn’t just read a natural-language instruction like “clean under the table.” It simultaneously processes camera frames, force feedback from the robot’s arms, and past interaction history in a single network, rather than stitching together separate perception and control stacks. Engineers say this reduces information loss and makes it easier to learn complex behaviors, from grasping a fragile glass to navigating a crowded hallway.

The approach also has strategic implications. By avoiding dependence on third-party models, the company keeps more control over performance, safety constraints, and data. In a climate of tightening export controls and rising concern about model provenance, that autonomy is part of the product.

Training on the job in Chinese homes

The robots are not staying in the lab. To accelerate learning, X Square Robot has launched a robotic cleaning service with home-services platform 58 Home, sending machines into real apartments under supervised conditions. Each deployment generates streams of “messy” data: unusual floor plans, cluttered surfaces, inconsistent lighting, and unpredictable human behavior.

Executives describe the model as “training upon deployment,” turning every job into a data point that flows back into WALL-A. It is a classic data flywheel: the more homes the robots visit, the more edge cases they see; the more edge cases they see, the more capable and marketable they become. For household customers, the promise is convenience; for the startup and its investors, it is a compounding advantage that is hard for latecomers to replicate.

The strategy also surfaces familiar tensions. Labor advocates in other markets have warned that cleaning and maintenance jobs could be automated away; privacy advocates worry about mobile cameras roaming through private spaces. Regulators, still grappling with text-based AI, will need new guardrails for robots that collect multimodal data inside homes and workplaces.

Beyond the home: factories, logistics, and care

X Square Robot’s roadmap reaches far beyond kitchen floors. The company plans to apply the same embodied intelligence stack to industrial manufacturing, logistics, and elderly care, seeking out environments where repetitive physical tasks and safety constraints dominate. In factories, a wheeled humanoid that can safely maneuver around people and adapt to new workflows without extensive reprogramming could appeal to manufacturers facing demographic and wage pressures.

In eldercare, the pitch is more delicate: assistive robots that can help with daily tasks or monitor for falls, without replacing human caregivers altogether. Policymakers and health systems will scrutinize these deployments closely, balancing cost savings and coverage against questions of dignity, reliability, and liability when machines fail.

China’s race for general-purpose robots

For China’s tech giants, embodied AI is emerging as a strategic arena alongside cloud, chips, and large language models. Xiaomi has already experimented with its own robots; backing X Square Robot lets it plug into a specialized team and a maturing platform. Alibaba, ByteDance, and Meituan bring distribution, data, and application scenarios, from e-commerce fulfillment to food delivery networks.

Whether X Square Robot can convert that backing into durable leadership is far from settled. The hardware remains fragile, the models are still prone to surprising errors, and the economics of large-scale deployment are unproven. Yet the size and speed of the company’s funding signal that in the eyes of China’s most powerful tech firms, the question is no longer whether embodied AI will matter, but who will own the stack when these machines finally move with confidence.