A Chinese embodied AI company just showed that the future of robotics might not require building a different AI system for every robot and task. At the 2026 World Robot Conference in Beijing, X Square Robot unveiled how its WALL-B foundation model powers multiple robotic platforms across home cleaning, logistics, and industrial applications, suggesting a more efficient path forward for the rapidly expanding physical AI industry.

What Makes This "One Model, Many Applications" Approach Different?

Rather than training separate AI systems for each robot type or task, X Square developed WALL-B as a general-purpose foundation model specifically designed for the physical world. Think of it like how large language models (LLMs) power different applications without needing to be completely retrained from scratch. At the conference, the same underlying model controlled wheeled humanoid robots performing household tasks, a five-fingered dexterous hand unboxing products, and a dual-arm system arranging flowers in vases.

The company demonstrated this flexibility across several real-world scenarios. In home environments, robots responded to spoken commands to organize items, retrieve objects, clean spaces, water plants, and handle pet-related chores. The system used coordinated two-arm manipulation and obstacle avoidance to move between rooms and complete requested tasks. This builds on X Square's earlier partnerships, including a March collaboration with Chinese local-services platform 58.com to launch robot-assisted home cleaning services, where robots worked alongside professional cleaners during household visits.

How Is X Square Proving This Works at Scale?

The most compelling evidence came from a five-hour logistics challenge during the conference. A WALL-B-powered system sorted 10,000 parcels in 5 hours, 14 minutes and 1 second, achieving an average throughput of 1,911 parcels per hour, or about 1.88 seconds per parcel. This wasn't a controlled lab test; the system handled parcels varying in size, shape, material, and position while maintaining consistency over thousands of items.

The logistics demonstration matters because it shows the model can handle real-world variability. The robots had to repeatedly identify, grasp, orient, and sort items while correcting unsuccessful actions and responding to changes on the conveyor belt. Maintaining that throughput over thousands of parcels demonstrates the consistency and endurance required for commercial operations.

Why Does This Matter for the Broader Robotics Industry?

The embodied AI sector is racing toward massive valuations, but most deployments remain limited to specific tasks or environments. X Square's approach challenges that fragmentation. According to the company's founder and CEO, the strategy reflects a fundamental belief about how embodied AI should develop:

"Embodied AI requires a foundation model built specifically for the physical world, in parallel with the foundation models developed for the digital world. At X Square, our aim is to develop a common intelligence foundation that can operate across different robots, tasks and physical environments, while continuing to improve through real-world deployment," said Wang Qian, founder and CEO of X Square Robot.

Wang Qian, Founder and CEO of X Square Robot

This philosophy aligns with broader trends in AI development. Just as foundation models in the digital world reduced the need to train separate systems for every language task, a foundation model for physical AI could accelerate deployment across industries. X Square is already operating WALL-B-powered systems on live parcel-sorting lines designed for continuous, round-the-clock operation.

Steps to Understanding How Foundation Models Transform Robotics

• Foundation Model Concept: A single large AI model trained on diverse data that can be adapted to multiple downstream tasks without complete retraining, reducing development time and costs for new applications.
• Real-World Deployment: Testing AI systems in actual homes, warehouses, and commercial environments rather than controlled labs, allowing the model to improve through exposure to genuine variability and edge cases.
• Multi-Modal Integration: Combining perception (vision and sensors), reasoning (decision-making), and action (physical movement) in a single system that can handle different robot morphologies and task types.
• Continuous Learning: Using data from real-world deployments to iteratively improve the foundation model, creating a feedback loop where more deployments lead to better performance.

The timing of X Square's demonstration is significant. The embodied AI market is attracting massive investment, with companies like Figure AI and others raising billions in recent months. However, many of these ventures focus on humanoid robots or single-task applications. X Square's emphasis on a generalizable foundation model suggests a different competitive strategy: become the underlying intelligence layer that powers diverse physical systems.

Meanwhile, the infrastructure supporting embodied AI is also advancing. STMicroelectronics and the National University of Singapore recently launched the ST-NUS HELIX Corporate Lab, a four-year research initiative focused on advancing edge AI hardware specifically for embodied intelligence applications. HELIX, which stands for Hardware for Embodied Low-power Intelligent Xcceleration, will develop memory-centric architectures and innovative in-memory computing to enable robots and other physical AI systems to process information efficiently on compact, power-constrained hardware.

The research will span the full technology stack, from AI models and system architecture to heterogeneous accelerators, on-chip memory hierarchies, circuit design, and silicon implementation. This represents the kind of foundational infrastructure investment that could make deploying embodied AI systems more practical and cost-effective across industries.

In the automotive sector, the convergence of autonomous driving and embodied AI is also creating new opportunities. Nio, the Chinese electric vehicle maker, reportedly plans to make a strategic investment in an embodied intelligence startup founded by its smart-driving chief Ren Shaoqing, while keeping him in his current management role. The new company has reached unicorn-level valuation, meaning it's valued at over $1 billion. This arrangement allows Nio to maintain stability in its autonomous driving team while staying connected to the rapidly developing embodied intelligence sector.

Smart driving and embodied intelligence share many foundational technologies, including perception, prediction, planning, control, and reinforcement learning. This technological overlap is drawing autonomous-driving talent into the robotics industry and making executives with mass-production experience attractive to startup investors.

X Square's approach suggests that the next phase of robotics development may not be defined by which company builds the most advanced humanoid, but rather which company develops the most capable and adaptable foundation model for the physical world. By demonstrating that a single model can power diverse robots across multiple industries, X Square is positioning itself as a potential infrastructure provider in an industry still figuring out its fundamental architecture.