Former Huawei Autonomous Driving AI Lead Huang Qingqiu Joins Moqi Intelligent Technology, Which Raises Over 1 Billion Yuan for General-Purpose Robots

Recently, Moqi Intelligence, an embodied intelligence company founded just six months ago, announced that it has completed over RMB 1 billion in angel round financing within six months, with a valuation exceeding RMB 7 billion. Gaorong Ventures participated in both consecutive rounds.

Moqi Intelligence was co-founded by Qingqiu Huang and Wenli Gao. CTO Qingqiu Huang, born in 1994, holds a bachelor's degree from Tsinghua University's Department of Automation and a PhD from The Chinese University of Hong Kong's MMLab, where he studied under Dahua Lin. He has long researched robot control, computer vision, and AI algorithms, publishing dozens of top conference papers. In 2020, he joined Huawei's Intelligent Automotive Solution Business Unit as a "Genius Youth" recruit, later becoming head of the autonomous driving AI department. He led algorithm breakthroughs and mass production for Huawei's intelligent driving systems from ADS 1.0 to 4.0, and was the first in the industry to achieve mass production at the million-unit scale for one-stage end-to-end architecture (WEWA).

CEO Wenli Gao was previously co-founder of cross-border logistics company iMile, and was the key figure in its comprehensive Middle East network rollout and replication of operating systems to global markets. Earlier, he spent 11 years at Huawei's carrier business line, responsible for overseas market expansion.

Moqi Intelligence is anchored in "systems engineering capability integrating hardware and software," using commercial service scenarios as training grounds for its technology, aiming for the super terminal of general-purpose home robots. In Huang's view, the endgame of embodied intelligence lies not in whose large model has more "spirituality," but in who can build a mass-producible, reusable, closed-loop underlying architecture in a chaotic, unstructured physical world.

Moqi Intelligence's endgame product will be a home robot with global physical understanding capabilities that can handle complex tasks — recognizing all changes in the home and autonomously completing cleaning, organizing, delivery, and various other tasks. As Huang put it, "I firmly believe that within ten years, general-purpose robots will become humanity's 'super terminal' in the physical world, entering millions of households."

The following is a conversation between An Yong Waves and Qingqiu Huang:

---

Building General-Purpose Robots That Truly Serve People

An Yong: New embodied intelligence companies keep emerging, and some share similar visions to yours. What do you think truly matters?

Huang: If I had to answer in one word: systems engineering.

In the lab, model architecture determines the ceiling; but in the real physical world at million-unit scale, systems determine survival.

Today the industry's hottest debates center on VLA versus world models, end-to-end versus hierarchical, autoregressive versus diffusion — but personally I don't think any of these are the most critical.

Embodied intelligence is fundamentally about building a real-time closed-loop system integrating hardware and software. What we truly need to do is start from first principles, fusing hardware boundary definition, rapid data flow, efficient model iteration, and closed-loop evaluation mechanisms into a self-evolving organism.

An Yong: What does a real-time closed-loop system mean?

Huang: Real-time and closed-loop are the essential characteristics of Physical AI. Large models "soliloquize" in static corpora; embodied intelligence "wrestles" in the dynamic physical world.

For example, when a robot lifts a paper cup full of water, it must adjust force every millisecond based on cup weight, water surface oscillation, and finger-to-cup friction — lifting without crushing or spilling. That's "real-time." And every action changes the environment, which immediately feeds back to affect the next action; action and environment are mutually causal. That's "closed-loop."

"Real-time" requires operation and motion control models to run on-device, with the "brain" at minimum 10Hz and the "cerebellum" at minimum 100Hz. This frequency redline pulls models from the cloud's "greenhouse of infinite compute" back to the on-device "cage of power consumption and thermal dissipation," in turn constraining compute and capping parameter counts — it's not that we don't want to scale up, but physical law forbids it. "Closed-loop" means any sensor precision loss or actuator latency gets amplified by the system, which also determines what models learn and how they learn. So at this stage, hardware and software simply cannot be decoupled for iterative development; the only path to rapid system iteration is hardware-software integration.

An Yong: This system is your underlying architectural logic. What about the product? What is Moqi Intelligence's ultimate product positioning and endgame direction?

Huang: We see the endgame in the home. Moqi Intelligence aims to build general-purpose home robots. We will release our first service-oriented robot in July, taking a step forward.

An Yong: A question many people have: compared to existing home appliances like robot vacuums and dishwashers, what is the necessity and core difference of general-purpose embodied robots entering homes?

Huang: These products are excellent, but they solve fragmented, point-based needs. And in solving problems, they often create new "human problems." Users must rearrange furniture, regularly clean dust bins and replace mop pads, load dishes according to the dishwasher's logic, set electronic boundaries for lawn mowers. This is humans serving machines, not machines serving humans.

Robots are the optimal form to change this status quo. Moqi Intelligence's endgame product will be a home robot with global physical understanding capabilities that can handle complex tasks. It can recognize all changes in the home and autonomously complete cleaning, organizing, delivery, and various other tasks. It requires no compromises from you; it adapts to your living order, not the other way around.

---

Autonomous Driving Is a Subset of Embodied Intelligence

An Yong: Why enter embodied intelligence now? Some embodied companies are already actively preparing for IPO — don't you feel it's late?

Huang: Not late at all. Now is precisely the moment when two technological inflection points converge.

The first is the GPT moment. Large language models have excellently solved high-level semantic understanding and task decomposition — robots finally have a "cognitive brain." But in my view, having only a brain is insufficient; the current bottleneck in embodied intelligence lies in low-dimensional physical execution. The second critical moment is the end-to-end moment in autonomous driving. I witnessed this history in the autonomous driving industry; it proved that using neural networks to drive real-time closed-loop control of a physical entity is feasible.

High-dimensional cognitive intelligence and low-dimensional physical control — these two originally parallel lines converge today, and their intersection is embodied intelligence.

As for companies preparing for IPO, that proves capital's long-term optimism for this track, but it doesn't mean technology and products have converged. Embodied intelligence is a systems engineering marathon; going public is just getting a supply coupon, the race has just started.

Moreover, from day one we've anchored on the general-purpose home robot endgame and pursued a hardware-software integrated approach, fundamentally different in underlying logic from existing players; home robots are a trillion-level global blue ocean, and we built our architecture for globalization from Day 1, so there's no "too late to enter." Getting the underlying system and product right matters more than rushing for short-term milestones.

An Yong: But many people see autonomous driving and embodied intelligence as two separate tracks. You come from autonomous driving — how do you view their relationship, and to what extent can autonomous driving experience transfer to embodied intelligence?

Huang: I think autonomous driving is a subset of embodied intelligence. Human operations in the physical world can be summarized as movement and contact. Autonomous driving handles two-dimensional planar movement, while embodied intelligence handles three-dimensional movement and contact. The shift from 2D to 3D means your solution space explodes — you need to contact objects, deeply understand the physical properties of various objects, which is also the greatest challenge of embodied intelligence relative to autonomous driving.

Autonomous driving, as a simplified version of embodied intelligence, has much experience worth drawing from. For example, end-to-end technical methodology, data closed-loop iteration systems, real-time on-device model optimization, multi-sensor fusion engineering solutions, even million-unit-scale mass production quality control processes — these are proven experiences from autonomous driving that can be borrowed in embodied intelligence R&D, substantially shortening the industry's cycle from demo to mass production.

An Yong: What difficulties do entrepreneurs in the industry generally underestimate?

Huang: First, data quality. Future embodied foundation models depend on scale-up-capable high-quality data. How to collect, filter, and process high-quality data is a systematic challenge.

Second, long-term stability of embodied systems. In unstructured worlds, environments are dynamic and full of perturbations. A system that "can run" doesn't mean it "can run at scale" — the gap between 1 and 1 million units shipped is a massive engineering chasm.

An Yong: Everyone is now collecting and training data through various paths. Is this the industry's cure?

Huang: Data is certainly most important, but it depends what data you're stockpiling. I firmly believe that the endgame form of embodied data must satisfy "three authentics": real operators, in real scenarios, doing real work.

The endgame should have two collection devices: one purely for video, extremely lightweight, perhaps just glasses, collecting massive video for pre-training; the other an improvement on existing portable devices, finding the sweet spot between wearability and reconstruction precision, collecting relatively smaller amounts of data for both pre-training and post-training — both must satisfy the "three authentics."

Additionally, data quality far outweighs quantity: getting sub-millisecond time synchronization, high-speed motion trajectory precision in weak-texture environments right, ensuring every hour of data is high-quality, matters far more than rapidly expanding to tens of millions of hours.

An Yong: What path do you follow for models?

Huang: Simplistically, a model is also a system, essentially doing two things: information compression and modality alignment. Embodied model inputs are high-dimensional tokens from images, touch, language at millions of dimensions, while outputs are action tokens at dozens of dimensions — both compression from high to low dimension and alignment across modalities. Low output dimensionality, sparse feedback, and large modality differences make it very difficult to learn.

What engineering must do is find ways to add "multimodal dense feedback." Today's heated debates over various approaches are essentially doing this in different dimensions: having models generate video and predict future frames is spatial dense feedback — the world model; having models express reasoning processes in language is logical dense feedback — an important training method for VLA; adding depth maps is geometric dense feedback, adding semantic segmentation is semantic dense feedback.

So world models and VLA aren't conflicting at all, not mutually exclusive, but "auxiliary lines" laid for models from different dimensions — which brings us back to that point: architecture debates are false propositions. Moqi adopts precisely a multi-expert architecture fusing various dense feedbacks, aiming to build a native multimodal model integrating "understanding — generation — decision-making."

An Yong: Developing proprietary native multimodal models and doing pre-training require sustained heavy investment. How do you evaluate the ROI?

Huang: Developing a general-purpose embodied brain is indeed long-term heavy investment, but I don't think this is fundamentally a short-term ROI calculation.

If we only stay at the body or single-task level, embodied intelligence can hardly bear the industry's current investment expectations. Long-term, true value will return to transferable brains, data closed-loops, and scalable product systems.

Of course, even if we were to calculate it, I think the numbers work out — model training investment is fixed cost after all, and for a high-value product shipping at massive volume, any fixed cost eventually becomes negligible.

---

"A Pragmatic Idealist"

An Yong: Was choosing entrepreneurship a difficult decision?

Huang: Building a great embodied intelligence company was a natural choice for me.

At Tsinghua, besides classes and sleep, I spent 80% of my time at Room 508 in the main building — Professor Mingguo Zhao's robotics lab. For four years there, I played with robots of every form, from line-following cars, self-balancing cars, to joining the Fire God team building humanoid robots, and finally for my thesis collaborating with seniors on an autonomous bicycle that could follow people. This is where I completed my embodied intelligence enlightenment.

I also met three seniors who, at that time, had just founded SenseTime with Professor Xiaogang Tang, as the first three employees. Through their referral, I interned at SenseTime. In the early days SenseTime had only about a dozen people, crammed into Wenjin International Apartments in Tsinghua Science Park — the living room was the office, bedrooms were dorms. I still miss that free, efficient, pursuit-of-extremes pure atmosphere. This atmosphere infected everyone there at the time, to the point that colleagues who pulled all-nighters writing code together went on to found numerous tech star companies like Momenta and MiniMax.

Having experienced that 0-to-1 fervor, the seed of "building a great tech company" in your heart can never be suppressed again.

An Yong: After your PhD you went directly to Huawei's Intelligent Automotive BU for autonomous driving algorithm R&D. What did Huawei mean to you?

Huang: If my PhD taught me to "look up at the stars and open new frontiers," five years at Huawei taught me to "keep my feet on the ground and fill the holes," giving me the most extreme iteration thinking and quality consciousness.

Huawei has a quality management system that could be called a "nightmare" — every minor issue is continuously tracked, and before every new version you face soul-searching questions: "Why wasn't this problem solved again?" The process is painful, but it's precisely this refusal to let any small problem go, grinding through version after version, that iterates excellent products. This accumulation gives me sufficient confidence today to build excellent embodied products, transforming fanciful technology into engineering answers that never fail in the physical world.

An Yong: If you didn't talk about your resume, how would you introduce yourself to a stranger?

Huang: I'm Qingqiu Huang, a pragmatic idealist.

I firmly believe that within ten years, general-purpose robots will become humanity's "super terminal" in the physical world, entering millions of households.

But at the same time, I'm even more clear that this day will never suddenly arrive because of one viral paper or one stunning demo. To cross the chasm from demo to product, there are no shortcuts — only trudging through mud pit by pit: obsessing over sub-millisecond multi-sensor time synchronization, scrutinizing every tiny error accumulation in motor actuators, cleaning every trajectory anomaly in training data, catching every exception branch from loose wiring or dropped packets. The road to the super terminal is destined to be long and arduous.

Source: An Yong Waves

By Ren Qian, edited by Ba Rui