Building a Foundation Stack for General-Purpose Robots

2026-07-13 · Source: IEEE Spectrum · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, long

Summary

Chinese embodied-AI company X Square Robot proposes an integrated stack for general-purpose robots, aiming to solve the lack of a "recipe" for transferable robotic intelligence. Their approach combines a data collection system, a world model, and an action model. The Universal Manipulation Interface (UMI) data system, QUANXTA Zero Series, focuses on data quality and cost, using wearable rigs and physical playback to validate interaction data, achieving comparable performance to all-robot datasets at a 20-fold lower cost. Its WALL-WM world model is organized around action-grounded semantic events, offering both variable-length event mode for reasoning and fixed-length mode for real-time control. The Wall-OSS-0.5 vision-language-action model is designed to be deployable on a real robot before fine-tuning, utilizing an X-Tokenizer for a semantic action interface that ensures capability transfer. X Square Robot's valuation has climbed above 20 billion yuan (US \$2.9 billion), reflecting investor confidence in their data infrastructure and foundation models. The company is releasing its world model code publicly for broader community validation.

Key takeaway

For robotics engineers developing general-purpose systems, X Square Robot's approach highlights critical design choices. You should prioritize data quality validation through physical playback, as smaller, cleaner datasets often outperform larger, noisier ones. Consider modeling robot behavior around action-grounded semantic events for improved long-horizon reasoning and ensure your pretrained models offer deployable capabilities before fine-tuning. This strategy can significantly reduce data collection costs and enhance model transferability across different robot embodiments.

Key insights

X Square Robot's integrated stack for general-purpose robots emphasizes high-quality interaction data, event-based world modeling, and pre-trained, deployable action models.

Principles

• The basic unit of robot data is an interaction, not a trajectory.
• Pretraining should yield usable capability, not just initialization.
• Behavior should be modeled around physical events, not fixed time slices.

Method

The UMI data collection system uses wearable rigs for human demonstrations, followed by physical playback on a real robot to validate task completion, ensuring high-quality interaction data.

In practice

• Use physical playback to validate robot interaction data quality.
• Model robot behavior around semantic events for long-horizon tasks.
• Design action models to be deployable before task-specific fine-tuning.

Topics

• Embodied AI
• General-Purpose Robots
• Robot Learning Data
• World Models
• Action Models
• Data Quality Control

Best for: Research Scientist, Investor, Entrepreneur, Robotics Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• This startup thinks robotics is about to have its ChatGPT moment — Video game data can train embodied AI foundation models, enabling rapid transfer learning for diverse robotic tasks.
• World Action Models give robots the ability to simulate consequences before they move — World Action Models enable robots to simulate action consequences, improving generalization and learning from unlabeled video data.
• Will Robotics Have a ChatGPT Moment? — General-purpose robotics will advance through coordinated AI systems and extensive real-world experience, not a single breakthrough.
• VIA: Visual Interface Agent for Robot Control — General-purpose foundation models can directly control robots through visual interfaces without specialized fine-tuning.
• Native Video-Action Pretraining for Generalizable Robot Control — LingBot-VA 2.0 is a purpose-built video-action foundation model for embodied robot control, enhancing generalization.
• How robots learn: A brief, contemporary history — AI models and large datasets are revolutionizing robot learning, enabling adaptation and complex task execution.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.