X Square Robot debuts foundation model for robotic butler after $100M Series A

The Quanta robot is intended to provide a path to humanoids in the household. Source: X Square Robot

Embodied artificial intelligence is drawing attention for its potential to enable robots such as humanoids to do more. X Square Robot today announced that it has raised an additional $100 million in Series A+ funding. The Shenzhen, China-based company also launched Wall-OSS, its proprietary open-source foundation model for robotic platforms, and the self-developed Quanta X2 robot.

“Contrary to popular belief, robots capable of reliably performing unpredictable real-world tasks, such as those in homes or hotels, are far from commercially viable,” noted X Square Robot. It said the key barriers are an over-reliance on task-specific training data and a disproportionate focus on bipedal locomotion.

Instead, the company asserted that there should be more generalized and reliable training in manipulation with robotic hands and fingers, as well as reasoning capabilities across diverse robot form factors.

“Training a robot to pick up a box in a warehouse does not prepare it to serve food on a cafeteria tray,” said X Square Robot. “This critical gap is precisely where the global rivalry to lead the future of AI is intensifying.”

Founded in 2023, X Square Robot said it is advancing humanoid robotics through training embodied AI foundation models. The startup has embedded its Wall-OSS platform in its products, including Quanta X1 and X2.

Robust, diverse training data key to unlocking embodied AI

X Square Robot claimed that its foundation model is “built upon the world’s largest embodied intelligence dataset.” The company has made the model open-source as it pursues its ambitious goal of making robotic butlers a reality within five years.

The Wall-OSS framework is capable of generalizing across multiple robot forms, said X Square Robot. It is designed to address two key challenges in current models. The first is that training embodied AI on new data overwrites previously learned knowledge, which the company called “catastrophic forgetting.”

The second challenge is modal decoupling, when vision, language, and action fall out of sync, explained X Square Robot. Wall-OSS is a multimodal embodied AI model trained on vision-language-action (VLA) samples, it said.

The training corpus for Wall-OSS combines real-world robotic action data with augmented generative video. The latter enables the model to learn under diverse environmental variables such as lighting conditions and object shape. This in turn strengthens its generalization and robustness across tasks, according to X Square Robot.

The company said this enables more effective training for fine motor control in humanoid hands and upper body movements. Its said its “’embodied brain’ is smart enough to not only independently handle unpredictable, non-routine real-world physical settings, but also continuously learn on the go.”

Wall-OSS uses FFN, three-stage training process

X Square Robot added that its model uses a shared attention + task-routed feed-forward network (FFN). This design lets Wall-OSS not only observe and analyze environments and objects, but also channel different types of information to specialized pathways at the same time.

“Since robots can’t absorb information at once the way humans do, the FFN directs each task to the most suitable pathway — like giving vision data, language instructions, or motor actions their own dedicated processors,” the company said. “These specialized pathways then work together, enabling the robot to complete complex or unfamiliar tasks more effectively.”

X Square Robot has trained Wall-OSS through a three-stage process that it said mirrors how planning and execution connect in the real world. It first learns high-level reasoning, like planning and task sequencing.

Then, the model practices fine-grained motor control, such as coordinating joints and trajectories. Finally, it fuses both, allowing the model to “think” through what needs to be done, but also carry it out smoothly in real-world environments, said X Square Robot.

On top of this, Wall-OSS integrates chain-of-thought (CoT) reasoning into its model, enabling robots to think many steps ahead before acting, even for scenarios that it has not been explicitly trained for, the company claimed. For instance, a robot can go from the dinner table to the fridge and then to the living room.

X Square Robot has developed the ArtiXon gripper and the X-OSS foundation model. Source: X Square Robot

Quanta aims to make robotic butlers a reality

Wall-OSS is optimized for maximum performance and is available for third-party robots and for developers through GitHub and Hugging Face. The open-source embodied AI framework is also integrated with Quanta X1, X Square Robot’s first-generation humanoid, and it will roll out alongside the new Quanta X2 humanoid.

Designed for both dexterity and versatility, Quanta X2 combines embodied AI with a robust mechanical design to enable applications across service, household, and industrial environments. It has a wheeled chassis and a robotic arm with seven degrees of freedom (DoF), with a total of up to 62 DoF throughout its body for flexible operation.

Each arm is equipped with a 20 DoF dexterous hand. The humanoid is capable of perceiving subtle pressure changes, creating a full-body teleoperation system that allows robots to communicate through lifelike gestures and emotional expression.

Quanta X2 includes a modular self-rotating tool clamp and can affix spinning brushes and mop head attachments for 360-degree cleaning.

X Square Robot to invest in making humanoids mainstream

New investors in X Square Robot’s Series A+ round included Alibaba Cloud, HongShan, and INCE Capital. The company also cited continued support from existing backers Meituan, Legend Star, and Legend Capital.

With its latest funding, X Square Robot said it plans to accelerate Wall-OSS adoption by expanding developer access, building out community-driven training datasets, and strengthening industry partnerships. It said it will also invest in the development and rollout of new humanoid models, broadening its portfolio beyond Quanta X2 to address a wider range of consumer and commercial applications.

“Together, these initiatives aim to push humanoid robots closer to mainstream adoption,” said the company.

Thanks in part to advances in AI, investors are bullish on humanoid robots. ABI Research predicted that the global market could reach $6.5 billion by the end of the decade, experiencing a compound annual growth rate (CAGR) of 138% between 2024 and 2030. Several exhibitors at IFA in Germany last month showed off potential household applications for humanoids.

Editor’s note: RoboBusiness 2025, which will be on Oct. 15 and 16 in Santa Clara, Calif., will include the Physical AI Forum and a track on humanoid robots. Registration is now open.

Keep track of your essentials with the Apple AirTag 4 Pack, the ultimate tracking solution for your belongings. With over 5,972 ratings and a stellar 4.7-star average, this product has quickly become a customer favorite. Over 10,000 units were purchased in the past month, solidifying its status as a highly rated Amazon Choice product.

For just $79.98, you can enjoy peace of mind knowing your items are always within reach. Order now for only $79.98 at Amazon!

Help Power Techcratic’s Future – Scan To Support

If Techcratic’s content and insights have helped you, consider giving back by supporting the platform with crypto. Every contribution makes a difference, whether it’s for high-quality content, server maintenance, or future updates. Techcratic is constantly evolving, and your support helps drive that progress.

As a solo operator who wears all the hats, creating content, managing the tech, and running the site, your support allows me to stay focused on delivering valuable resources. Your support keeps everything running smoothly and enables me to continue creating the content you love. I’m deeply grateful for your support, it truly means the world to me! Thank you!