A Chinese robotics startup wants to build the 'recipe' that gives robots general intelligence
X Square Robot is open-sourcing an AI stack that aims to do for physical robots what large language models did for text. The approach is interesting. The independent evidence is still thin.
Key points
- X Square Robot, a Chinese embodied-AI company, is releasing its full robotics AI stack as open-source software in 2025.
- The company claims its data-collection method reaches performance comparable to all-robot datasets at roughly 20 times lower cost.
- Its world model, WALL-WM, organises robot behaviour around named actions like "reach" or "grasp" rather than fixed time windows.
- Its action model, Wall-OSS-0.5, is designed to run on a real robot before task-specific training begins, which is rare in the field.
- All headline results so far come from the company's own benchmarks, meaning independent testing hasn't yet confirmed them.
Teach an AI system enough text, and something useful emerges. That insight, called pretraining on broad data, turned large language models (the technology behind chatbots like ChatGPT) into general tools that millions of people use every day.
Robotics has no equivalent shortcut. A robot that folds laundry can't simply transfer that skill to loading a dishwasher. The systems inside most robots today are built from separate, specialist parts that don't add up to anything general.
X Square Robot, a Chinese company focused on what researchers call embodied AI (AI that lives in a physical body and must act on the real world), thinks it's found the recipe. Unusually for a Chinese AI startup, it's releasing the core of it for anyone to use and inspect.
The company's approach rests on three layers working together: how robots collect learning data, a world model that predicts what will happen next in the physical world, and an action model that turns perception and reasoning into actual movement.
What makes this different from other robot AI projects?
The most striking piece is how X Square Robot handles training data. Collecting robot demonstrations is expensive because you need a real robot, careful operator supervision, and hours of setup. X Square Robot built a wearable rig with two grippers that a person wears on their hands, no robot required, cutting collection costs sharply.
But cheaper data is worthless if it's wrong. The company adds a quality check that's genuinely unusual: a sample of recorded movements is physically replayed on a real robot, and only movements that complete the task are counted as valid. A gripper closing a fraction too early might look like a successful grasp in the recording, but the object has already been knocked away. That trajectory gets discarded.
Combining this cheaper human-demonstration data with a small amount of real-robot data reportedly reaches performance comparable to datasets collected entirely on robots, at about one-twentieth of the cost. Striking, if true. As IEEE Spectrum AI noted, these numbers come from X Square Robot's own tests, and outside researchers haven't confirmed them yet. We first reported on simulation-first robot training on 13 July 2026 in Train the Robot in a Fake World First, Then Send It Into the Real One, and the validation problem flagged there applies here too: lab numbers don't always survive contact with other people's hardware.
How does the world model work?
WALL-WM takes a different approach to time. Most robot AI systems slice movement into equal-length chunks, the way a film is cut into frames. WALL-WM instead organises behaviour around meaningful events: reach, grasp, place. Each event ends when the action completes, not when a clock ticks over. Boundaries match what's actually happening.
The action model, Wall-OSS-0.5, is a vision-language-action model (software that reads camera images and language instructions, then produces physical movement). The company's requirement is that it must do something useful on a real robot before anyone trains it further for a specific task. General ability first, specialisation second.
Should you expect a home robot soon?
Not yet. General-purpose home robots remain years away regardless of this announcement. What matters here isn't the performance numbers, which are self-reported and narrow in scope. It's the open-source release. If independent researchers can reproduce and extend these results, the field moves faster than it would under a closed, proprietary system. That's the real bet X Square Robot is making, and it's worth watching whether the broader robotics community picks it up.
