Build one general robot brain that can be dropped into many robot bodies and handle everyday tasks, starting with household chores, logistics sorting and cleaning.
WALL-B unveiled Apr 2026; robots on trial in homes and a cleaning service in Shenzhen and Beijing.
Updated 21 Apr 2026Checked 11 Oct0 updates this week
WALL-OSS builds on a 3-billion-parameter model that already understands images and text, then is taught to output robot motions.
One network can go from an instruction to a written reasoning step, a list of subtasks and then smooth joint motions, or skip straight to motion.
Training first predicts compressed action tokens, then a flow-matching head that generates continuous, smooth arm trajectories.
WALL-B learns vision, language and action together from real-home data, giving the robot an intuitive sense of how objects react to its moves.
| Spec | WALL robot foundation models |
|---|---|
| WALL-OSS base model | Qwen2.5-VL, 3B parametersR (reported) |
| WALL-OSS robot training data | Over 10,000 hoursR (reported) |
| WALL-OSS-0.5 autonomous completion | Over 80% on 4 of 17 real-robot tasks, untunedR (reported) |
| Quanta X2 degrees of freedom | 62R (reported) |
R reported by the company
X Square Robot
WALL robot AI models, Quanta wheeled robots
X Square Robot is a Shenzhen start-up, founded in December 2023, that builds AI models to control general-purpose robots, plus its own wheeled two-armed Quanta robots and dexterous hands. It open-sourced its WALL-OSS model.
