The embodied AI start-up says its new foundational model will end an era of ‘scripted' robotics with the entry of robots to real homes within 35 days

image credit: Bamboo Works

General purpose embodied intelligence startup X Square Robot on Tuesday unveiled a new embodied AI foundational model aimed at accelerating robot deployment in homes, addressing long-standing concerns about the gap between demonstrations and real-world performance.

The company's Wall-B foundational model is built on its proprietary World Unified Model (WUM) architecture, which includes unique features such as native multimodality, a "world view" of the physical world and the ability to evolve based on interactions with the real world.

The model marks a shift in how robots are trained. Traditional systems typically rely on modular architectures, where vision, language and motion modules are developed separately and later stitched together. That approach has worked well in structured environments like factories, but tends to break down in variable environments like homes.

Wall-B works differently by jointly training for perception, language, action and physical prediction from the outset, creating a single system that learns to interpret and interact with external environments in an integrated way.

"Robots in factories … repeat the same action 10,000 times without variation. In a home, however, they need to perform 10,000 different actions, each unique and non-repetitive," said X Square Robot founder Wang Qian. "Therefore, the challenge of a truly intelligent robot lies not in repeating a single action, but in the ability to execute new, untrained movements within unstructured environments."

Unlike many robotics programs that rely heavily on synthetic data or carefully curated datasets, Wall-B is trained on data from real, lived-in environments that robots often struggle with, such as cluttered rooms and unpredictable human activity. That exposes the model to the "noise" of everyday life early on, improving a robot's ability to generalize.

The system also incorporates a physics-aware prediction mechanism that allows it to simulate the results of actions before executing them. X Square likens the process to how humans catch a ball: rather than waiting for the ball to arrive, the brain predicts its trajectory the moment it is thrown. Its model applies similar predictive reasoning, allowing robots to account for gravity, friction, and collision risk in real time.

Together, these features aim to address a persistent challenge in robotics: the gap between simulated training environments and real-world deployment – an essential capability for safe deployment of robots in unpredictable home environments.

"This is a countdown, not a concept," said Wang. "The model is already operating in real-world settings and improving continuously, with human intervention expected to decline over time."

The company said it plans to begin placing robots into ordinary households within 35 days, an aggressive timeline in a sector where commercialization has repeatedly lagged technical progress.

The foundation model is designed to integrate closely with the company's hardware, including the QUANTA X1 Pro dual-arm wheeled robot and the more humanoid QUANTA X2, both equipped with highly dexterous hands.

X Square Robot was founded in 2023 in Shenzhen, and has quickly emerged as one of China's leading embodied AI startups, using an end-to-end approach to robotics and backed by billions of yuan from investors including Alibaba, Meituan, Xiaomi and ByteDance.

Benzinga Disclaimer: This article is from an unpaid external contributor. It does not represent Benzinga’s reporting and has not been edited for content or accuracy.