X Square Robot is using the World Robot Conference 2026 in Beijing to make a broader argument about the future of physical AI: general-purpose robots will depend as much on data, foundation models and reusable skills as they do on hardware. The company’s demonstrations connect those layers—from embodied-data generation and dexterous manipulation to logistics sorting, flower arranging and household tasks—showing how AI models are beginning to move from controlled demonstrations into messy, variable real-world environments.

The next phase of artificial intelligence may not happen inside a chatbot window. It may happen on a warehouse floor, in a factory or inside a home.

That is the proposition behind X Square Robot’s exhibition at the World Robot Conference 2026 in Beijing, where the Chinese embodied AI company is demonstrating a technology stack spanning robot data collection, foundation-model development, manipulation skills and physical deployment.

At Booth C107, the company is presenting its WALL-B embodied AI model alongside six-axis robotic arms in a logistics-sorting demonstration. The system is designed to identify parcels, pick them up, reorient them and feed them into downstream processes. Unlike conventional automation built around relatively predictable objects, the demonstration includes parcels with different sizes, weights, materials and shapes, including soft packages that require labels to be flattened.

X Square Robot says the system reached 1,816 parcels per hour with more than 98% accuracy during a livestreamed demonstration on August 12. Those figures come from the company rather than an independent benchmark, but the underlying challenge is significant: industrial robots need to handle variation rather than simply repeat a perfectly choreographed motion.

That distinction is becoming central to the physical AI race.

The International Federation of Robotics reported that 542,000 industrial robots were installed globally in 2024, more than twice the level of a decade earlier. China alone accounted for 54% of global installations, with 295,000 units deployed during the year.

The next competitive layer is therefore shifting from simply putting more robots into factories to making those robots adaptable.

From robot hardware to foundation models

X Square Robot’s approach resembles the emerging architecture used by leading AI and robotics companies: separate the physical machine from the intelligence, data and training infrastructure that make it useful.

Its QUANXTA Zero platform is intended to generate embodied training data without requiring a physical robot for every data-collection task. The company says the platform covers data processing, annotation, training and evaluation.

That matters because physical AI has a data problem. A language model can learn from enormous quantities of digital text, while robots need information about objects, motion, spatial relationships, forces and physical consequences. Gathering that information in the real world can be expensive and slow.

Competitors are attacking the same bottleneck from different directions. NVIDIA’s Isaac GR00T combines robot foundation models, data pipelines, simulation, middleware and edge computing for humanoid development. NVIDIA describes the architecture as a path from model training and simulation to real-time deployment on robots.

Google DeepMind has taken a model-centric approach with Gemini Robotics, including vision-language-action models that translate visual information and instructions into actions for robots. Its on-device model is designed to run locally and adapt to tasks without depending entirely on cloud inference.

X Square Robot is positioning itself in this same emerging category, but with a particularly broad emphasis on the connection between embodied data, foundation models, manipulation skills and commercial robot deployments.

Teaching robots to handle the unexpected

The company’s flower-arranging demonstration illustrates why this approach matters beyond logistics.

A visitor can give a natural-language instruction specifying a preference, such as a particular rose color. The robot must interpret the instruction, identify the relevant object, manipulate it and execute a sequence of actions. It can also adapt when flower order or object positions change.

That is closer to an AI agent interacting with a physical environment than traditional industrial automation.

The challenge is not merely recognizing an object. A useful general-purpose robot needs to understand an instruction, perceive the current state of its surroundings, plan a sequence and continuously adjust its movements as conditions change.

NVIDIA’s research similarly describes robot foundation models as a way to generalize skills such as grasping, moving and transferring objects across tasks and robot embodiments.

X Square Robot is also demonstrating a dexterous-hand platform focused on skills such as grasping, twisting, opening and tool use. The company’s fan-handling example highlights another important frontier: fine manipulation is considerably harder to generalize than simple pick-and-place operations.

Bringing physical AI into the home

The most ambitious part of X Square Robot’s demonstration moves beyond industrial environments.

Its X Family Member Program recreates a household and organizes robot interactions around everyday activities including meals, entertainment, leaving home and remote interaction through an app. The company says it also placed robots in real households for extended interactions beginning in May.

X Square Robot has separately partnered with 58.com to provide paid home-cleaning services in China, suggesting that the company is testing whether embodied AI can move from laboratory demonstrations into service-based business models.

That transition is important because service robotics is already expanding. The International Federation of Robotics says nearly 200,000 professional service robots were sold in 2024, up 9%, while transportation and logistics represented more than half of professional service-robot sales.

The commercial opportunity is therefore not hypothetical. What remains uncertain is whether general-purpose robots can achieve the reliability, safety and economics required for widespread deployment.

The enterprise AI lesson

For enterprise technology buyers, X Square Robot’s strategy points to a broader shift in AI infrastructure.

McKinsey’s 2025 global AI survey found that 62% of respondents said their organizations were at least experimenting with AI agents, although most companies remained early in scaling them across the enterprise.

Physical AI introduces another layer of complexity. An AI agent operating software can generate an incorrect answer; a robot acting incorrectly can damage equipment, products or people.

That makes the infrastructure underneath the model critical. Simulation, evaluation, telemetry, edge inference, training data and safety controls become part of the AI stack rather than peripheral robotics components.

For companies evaluating embodied AI, the important question may therefore be less which robot is smartest? and more which platform can continuously improve the robot once it enters the real world?

X Square Robot’s WRC 2026 exhibit is effectively an answer to that question. Its bet is that general-purpose robotics will emerge from an integrated loop connecting data, foundation models, reusable skills and physical machines.

The industry is increasingly converging on that architecture. The open question is how quickly those systems can move from impressive demonstrations to dependable, economically viable workers in environments that were never designed for robots.

Market Landscape

The embodied AI market is evolving from conventional industrial automation toward AI-native robotics, where foundation models and data pipelines determine how flexibly machines can perform new tasks.

Three competitive approaches are emerging:

• NVIDIA is building a broad development stack around Isaac GR00T, simulation, data pipelines and Jetson edge computing.
• Google DeepMind is developing vision-language-action models through Gemini Robotics, emphasizing multimodal reasoning and task execution.
• Amazon is applying AI to a huge deployed robotics fleet, reporting more than one million robots in its operations and using its DeepFleet system to optimize robot movement.
• X Square Robot is emphasizing an integrated path from embodied-data production and foundation models to industrial and household applications.

The market’s scale is already substantial. Global industrial robot installations reached 542,000 in 2024, while professional service-robot sales approached 200,000 units.

The next phase will likely be measured less by unit shipments alone and more by task generalization, deployment economics, autonomy and the ability to train robots continuously from real-world experience.

Top Insights

• X Square Robot is connecting embodied data, foundation models and physical machines, giving enterprises a potential path toward more adaptable AI-driven automation.
• Its WALL-B logistics demonstration targets variable parcels, showing why perception and manipulation increasingly matter alongside conventional robotic speed and precision.
• QUANXTA Zero addresses the data bottleneck by supporting embodied-data production without requiring physical robots for every training and evaluation workflow.
• Google DeepMind and NVIDIA are pursuing similar physical-AI infrastructure, making foundation models, simulation and reusable robot skills increasingly important competitive layers.
• For enterprises, the long-term opportunity is general-purpose automation across logistics, manufacturing and homes, but reliability and safety remain critical adoption barriers.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI