WAIC Robot Exhibition Highlights: I Was Captivated by the Most Stylish 'Moz1'}

During the annual WAIC, we were captivated by the lively and versatile humanoid robots on display.

As 2025 progresses, frontiers in AI continue to break boundaries, with embodied intelligence (Embodied AI) and its carriers gaining significant attention domestically, especially humanoid robots.

From the dance routines on the Spring Festival Gala stage at the start of the year to the first global humanoid robot marathon and fighting competitions held in April and May, advances in AI algorithms and machine learning have made these robots increasingly agile, adaptable, and capable of perception, decision-making, and action in complex environments.

At WAIC 2025, the embodied intelligence zone, including various humanoid robots, was undoubtedly the highlight. These robots, with their diverse forms and dynamic movements, drew crowds eager to see their skills firsthand.

One booth, in particular, caught our attention: a humanoid robot acting like a human waiter, taking orders, retrieving drinks from the fridge, and handing them to visitors.

We observed the entire process—from recognizing commands, locating the target, to grasping and delivering drinks—executed seamlessly through autonomous reasoning by the VLA model, showcasing astonishing intelligence.

Further investigation revealed that this is developed by Qianxun Intelligence, a startup founded in February last year, dedicated to creating general humanoid robots and next-generation embodied large models to build a new era of intelligent workforce.

The robot showcased at their booth, "Moz1", was officially launched in June. It is the first domestically developed high-precision full-body force-controlled embodied robot, with 26 degrees of freedom (excluding dexterous hands). Its integrated joints have 15% higher power density than Tesla's Optimus, achieving industry-leading speed, accuracy, safety, and biomimetic force control.

What other surprises does Moz1 hold? At the Qianxun booth, we saw many more impressive skills.

Humanoid Robot Moz1 "Stuns" WAIC Audience

At WAIC, the impact of these robots was felt almost constantly.

We first visited the motion control zone of Moz1.

Here, Moz1 demonstrated excellent coordination, dynamic balance, stability, and intelligent planning capabilities.

Watch as Moz1 performs spacewalk-like movements, with fluid motion and no spillage when holding a water cup:

It also performs extreme S-shaped bends effortlessly:

This clearly shows a qualitative leap in humanoid robots' ability to adapt to dynamic environments.

Next, we moved to the remote operation zone, where Moz1’s human-robot collaboration skills were on full display.

Under the guidance of technicians, Moz1 played a mini-maze game:

It also stacked blocks:

And danced:

This indicates Moz1’s exceptional flexibility and precision in fine task operations.

Finally, in the clothing folding zone, it handled flexible objects with ease—grasping, folding, and stacking clothes into perfect blocks:

After experiencing Moz1, the biggest takeaway is that robots are no longer just executing preset tasks—they can now make intelligent decisions, self-optimize, and adapt based on environmental changes, greatly improving their flexibility and efficiency in real-world scenarios.

In short, the evolution speed of humanoid robots is surpassing our imagination. Behind all this is Qianxun Intelligence’s full-stack in-house embodied robot technology and systematic hardware-software integration.

Building a "Brain-Body" Unified Robot Framework

From its inception, Qianxun Intelligence aimed to develop both the next-generation embodied large models and general-purpose humanoid robots, with the brain and body evolving together. The large model handles learning and decision-making, while the robot body handles perception and execution, with both advancing in tandem to create truly generalizable, deployable robots.

At the model level, they adopted an end-to-end VLA (Vision-Language-Action) architecture, a mainstream approach in recent embodied large models, exemplified by frameworks like Helix and Redwood AI. This approach connects perception, understanding, and action, accelerating the development of versatile embodied intelligent agents capable of generalization and complex task execution.

Qianxun targeted this practical, scalable framework and released their in-house VLA model Spirit v1 in March, which has achieved significant leaps in adapting to complex environments and diverse tasks. As showcased at WAIC, the AI-powered robotic arm enabled by Spirit v1 performed the full process of folding clothes smoothly, solving a major industry challenge in long-range flexible object manipulation.

Spirit v1 continues to evolve, exploring more real-world applications and robustness. The recently launched Moz1 robot, equipped with this model, marks a key milestone, demonstrating the system’s readiness for deployment in real scenarios.

With its strong generalization, Moz1 can accurately perform tasks like desktop tidying, trash disposal, seat adjustment, and blackboard cleaning—becoming a versatile "jack-of-all-trades" in office environments:

It can also organize desktops:

Dispose of trash:

Adjust chairs:

And clean blackboards:

As Moz1 accumulates more real-world interaction experience, it will lay the foundation for more complex collaborative tasks. In the future, it may evolve from "capable worker" to "thinking assistant," becoming an all-in-one office helper—no longer just a dream.

               Qianxun’s all-in-one capabilities include software and hardware breakthroughs, supporting its core mission.

In addition to the software innovations, Moz1 has achieved multiple hardware breakthroughs, including:

• Full-body configuration with 26 degrees of freedom, enabling high flexibility and multi-pose tasks;
• Equipped with the world’s highest power density integrated force-controlled joints, delivering high output, responsiveness, and lightweight design;
• Leading domestic high-precision, high-speed WBC (Whole-Body Control) system for complex motion planning;
• Load-to-weight ratio of 1:1, allowing it to handle loads comparable to its own weight, enhancing practicality.

Combined with its powerful, deeply integrated brain and body, Moz1 achieves multimodal perception, full-body coordination, and real-time remote control, forming a data feedback loop for continuous learning and optimization.

This full-stack optimization—from algorithms to hardware—makes Moz1 a top-tier robot. The deep synergy of AI model evolution and hardware breakthroughs forms the core competitive advantage of Qianxun Intelligence in embodied AI.

Spirit v1 and Moz1 are the results of over a year of innovation. Moving forward, Qianxun will further explore the deep coupling of VLA models and robot bodies in complex environments, enhancing system generalization and robustness. They will also focus on demand-driven commercial deployment across manufacturing, service, and other core scenarios, transforming technological breakthroughs into market-ready products.

This comprehensive approach aligns with industry needs for highly flexible, adaptable, and intelligent robots, paving the way for broader applications and smarter, cross-scenario deployment.

Continued Investment: Not Just Technology, But the Future

Since ChatGPT’s debut, capital has been seeking the next big thing. Embodied robots capable of understanding, reasoning, and executing tasks are believed to trigger the next AI wave.

In recent years, domestic investments have surged into companies like Yushu Robot, Qianxun Intelligence, Zhiyuan Robot, and Xingdong Jiyuan, entering a period of concentrated funding.

Qianxun Intelligence, less than 18 months old, has secured multiple rounds of funding, including a recent Pre-A+ round of nearly 600 million RMB led by JD.com, with participation from China Merchants Venture, Zhejiang Science and Technology Innovation Fund, Huatai Zijin, and Fosun Ruilian. Existing investors like Shunwei Capital and Huaguan Fund also increased their stakes.

The high valuation and rapid growth of these companies reflect the industry’s optimism about embodied AI, with Goldman Sachs predicting the global humanoid robot market could reach $154 billion by 2035, opening vast opportunities for Chinese and international players.

Recently, Nvidia CEO Jensen Huang expressed optimism about China’s robotics industry during the third Chain Expo, citing China’s strengths in AI, mechatronics, and manufacturing as key advantages fueling rapid growth in embodied intelligence.

With the deep integration of large models and robotics, local companies like Qianxun are building a complete ecosystem from core technology to end-user products. Led by founders Han Fengtao, Gao Yang, and Zheng Lingyin, with extensive global market experience, they are creating an efficient growth path through full-chain development—technology, product, and market validation.

We look forward to seeing what new innovations humanoid robots will bring in the near future.