ZhenFund angel project "YuanCe Future" (源策未来) has raised hundreds of millions of RMB in its Series A funding round, aiming to build the OpenAI of the humanoid robot era.

Recently, Archon Robotics, a company developing full-body humanoid foundation models, announced the completion of a seed round of several hundred million RMB. Investors include leading USD funds such as ZhenFund, Gaorong Ventures, IDG Capital, and 5Y Capital, as well as diversified support from institutions including Gobi Partners' joint fund with The University of Hong Kong, MiraclePlus, and Shanghai Academy of AI for Science. Lighthouse Capital serves as Archon Robotics' long-term exclusive financial advisor. The funds will primarily be used for R&D of full-body humanoid foundation models, multimodal full-body motion data collection, talent team expansion, and the establishment of multi-location R&D centers and industrial partnership ecosystems — accelerating the open-source release of its humanoid base model within this year.

Qin Tianyi, Managing Director at ZhenFund, commented: "We've known Hongyang Li for many years, and we've watched him lead his team to tackle the hardest and most important problems — from UniAD to BEVFormer, and now to Archon's vision of a general-purpose full-body humanoid foundation model. The team never chases benchmarks; they're resolutely pushing the boundaries of intelligence. What's even more remarkable is that the OpenDriveLab Hongyang founded has become a true talent valley for autonomous driving and embodied AI, cultivating a cohort of fiercely motivated top young researchers. Archon co-founders Tianyu Li and Li Chen are outstanding representatives of this group."

Archon Robotics was founded in April 2026, with its R&D headquarters in Shanghai's Caohejing Development Zone, Xuhui District. The company focuses on developing general-purpose full-body humanoid foundation models, building Whole-body Intelligence to equip humanoid robots with human-like full-body mobility and manipulation capabilities — accelerating the path toward embodied AI entering millions of households.

Founder Dr. Hongyang Li is currently an Assistant Professor at The University of Hong Kong, Assistant Dean at the School of Computing and Data Science, and a mentor at Shanghai Academy of AI for Science. His end-to-end autonomous driving project UniAD won the Best Paper Award at CVPR 2023 — the only such honor for a mainland Chinese academic institution in nearly a decade. In 2026, he received the RSS Early Career Award, becoming the first Chinese scholar in the award's 20-year history.

Co-founder and CEO Dr. Tianyu Li is among the first graduates of Shanghai Academy of AI for Science and holds a PhD from Fudan University. He was a core developer of Huawei's mass-production autonomous driving ADS 4.0 world engine solution, and is an outstanding young researcher in end-to-end autonomous driving and embodied AI — selected among the EAI Academic Rising Stars 20 by Alibaba Cloud's ModelScope community in 2025. Co-founder and Head of AI Dr. Li Chen graduated from Shanghai Jiao Tong University's Zhiyuan Honors Program and received the HKU Presidential PhD Scholarship. As first author of the UniAD best paper, he is an accomplished young researcher in embodied AI and world models, recipient of the 2024 WAIC Rising Star Award, and has been invited to serve as Area Chair for ECCV 2026.

Core team members hail from leading institutions including The University of Hong Kong, Tsinghua University, Shanghai Jiao Tong University, Fudan University, and Zhejiang University, with deep expertise in top autonomous driving, robotics, and large model research teams. They've worked together for over five years, long transcending the typical teacher-student relationship, forging extreme combat-level默契 through countless technical battles.

The Archon team toasting at Everest Base Camp, February 2024. The steeper the peak, the more extreme默契 required.

A team with such a strong autonomous driving background — what gives them the confidence to rapidly produce top-tier work in humanoid robotics?

Archon's answer is clear: high-level autonomous driving and humanoid robotics are fundamentally同源 at the底层逻辑 level. Archon's team combines cutting-edge original algorithm breakthrough capabilities with massive-scale industrial system deployment experience, enabling them to directly translate mature complex systems engineering prowess into evolutionary speed for robotic brains.

Archon well understands that migrating from wheeled chassis + dual arms to humanoid form involves no painless transition. While many top overseas companies are only beginning to grasp the severity and complexity of humanoid tasks, Archon has long anchored its North Star.

The embodied AI industry is entering a critical moment of divergence. Public data from Omdia and others shows that in the first half of 2026, China's embodied intelligence and robotics sector saw 288 financing events, with disclosed funding totaling over 46 billion RMB — rapidly approaching the 55.4 billion RMB scale for all of 2025. Yet massive funding has not brought convergent technical consensus.

Most current embodied solutions suffer from inherent limitations: existing training data consists primarily of desktop first-person video, single-arm or gripper movements — missing human-native interaction logic such as whole-body center-of-gravity adjustment, torso leverage, and multi-limb coordination. This means most robots can only perform fixed-point grasping, struggling to autonomously adapt to variable tasks like opening doors, making beds, or two-handed object manipulation that require full-body cooperation.

The root of this limitation lies in structural gaps in data infrastructure. CEO Tianyu Li believes "the embodied datasets available on the market appear vast, but the information truly effective for full-body humanoid training is extremely limited."

First-person video datasets only capture what the human eye sees, missing critical motion pose information beyond hand appearances — squatting, bending, leaning sideways. Robotic arm and gripper annotation data is largely confined to planar scope, recording only end-effector trajectories: models can learn how manipulators move, but not how to interact with environments. Meanwhile, real humanoid robot data is already scarce, with collection costs running from hundreds to nearly a thousand RMB per hour, and samples of whole-body multi-joint coupled compound tasks have been virtually absent from past data pools.

These three data types each have their gaps, but point to the same problem: the core information of how humans complete daily actions — how the whole body coordinates, how center of gravity shifts, how force transfers from lower to upper limbs — has almost never been recorded in existing data.

"The long-term absence of this information locks current robot capabilities at the level of fixed desktop grasping, with a data chasm separating them from the diverse tasks of real home environments," Li said. "To break through this ceiling, we must return to the source and redefine the logic of data collection."

Therefore, Archon Robotics has chosen a domain almost no one has entered: building a general-purpose full-body humanoid foundation model. Its core concept, Human Body Learning, aims to learn human whole-body poses and coordination patterns, rather than merely tracking end-effector trajectories. This allows robots to acquire "the wisdom of limb coordination" from human full-body movements, thereby possessing complete full-body interaction capabilities.

By沉淀ing humanoid robot action intelligence as much as possible at the "midbrain" level independent of any specific body, the capabilities learned by this midbrain are no longer tied to particular robot models. It outputs whole-body motion trajectories rather than joint-angle commands for specific models. This gives the model potential to migrate across different embodiments. As data collection becomes more comprehensive and scene coverage more diverse, the midbrain's representational power grows stronger, and the range of body types to which Archon Robotics' embodied full-body brain can migrate expands.

Based on this judgment, Archon will build an entirely new data collection system. Founder Hongyang Li believes the evolution path of embodied data is progressing from real-machine teleoperation toward handheld devices and first-person perspectives, with the ultimate destination being human-centric full-humanoid data with complete human perceptual elements and full-body motion labels.

This is not a comfortable path. Motion capture costs are extremely high, and environmental requirements are stringent. Recalling technical battles in uncharted territory, a core Archon R&D member candidly shared: "Doing something unprecedented — the biggest feeling is that if there's even a trace of doubt in your subconscious that it can't be done, it probably never will be. Archon's only faith is unconditional belief, then charging forward like a pawn crossing the river, cutting down every monster that emerges one by one."

Meanwhile, Archon Robotics will also introduce multidimensional perceptual modalities such as touch, paired with higher-precision full-body and hand capture equipment. Li believes data diversity and quality matter more than raw scale. "A single piece of whole-body data covering center-of-gravity movement and torso angle changes carries far higher information density than a hundred pieces of desktop data with only hand trajectories."

How data is collected determines what models can learn; a model's capability gaps in turn define the next collection target. Once this "collect-train-feedback" loop is operational, it forms a continuously self-reinforcing data moat: with each round of collection and training, model capabilities improve, the system's understanding of "which data is truly useful" becomes more precise, and the efficiency and quality of the next round of collection rise another level.

It tests not only algorithmic engineering capability, but systematic understanding of the fundamental question "what exactly does the model need to learn from the physical world." And this understanding is precisely Archon Robotics' core judgment.

Archon plans to release its first humanoid-native foundation model in late 2026. For humanoid robots to truly move from laboratories into homes, what's needed is not a single perfect point demonstration, but the ability to work continuously and reliably in complex, dynamic, unstructured home environments. The ceiling of this capability fundamentally depends on the depth of the model's understanding of the physical world.

Archon chooses to return to the starting point of embodied intelligence to answer this question anew: what kind of body, and what kind of data to learn from, determines how far robots can ultimately go. Just as there's no helicopter to the summit of Everest, there are no painless shortcuts to general embodied intelligence. With the most primal reverence for the physical world and the most audacious technical ambition, Archon is launching a charge with no retreat toward the ultimate intelligence of the physical world.

Moonshot AI｜Hero Games｜Biansai Technology

RightBrain AI｜Vastech Energy｜UZIS

Helium Starlight｜Yahaha｜SmartMore Perfect Diary｜Jingying Technology｜Hesai Technology

Youibot｜Zero Zero Robotics｜CoEvolution NuMiao｜KnowYourself｜Manifold Tech Mogic AI｜Xiaohongshu｜Black Lake｜YuanGu

Yuguang Technology｜StarDetect｜Qingyue Technology CellX｜Yunzhou-Tech｜GuNa Technology Di-Matrix｜Weiming Shiguang｜Xellar Biosystems