A Brilliant Vision: The Rapid Rise from a Tsinghua Lab to a ”Unicorn” in Physical AGI

trade2dys agoupdate AiFun
26 0

3 monthsfinancing3.5 billion, with a valuation exceeding 10 billion—China’s answer to the global model industry has emerged

On June 15, 2026, Jijia Vision secured another 1 billion yuan in Series B2 funding. Investors included Singapore’s Lion City Capital, the China-Belgium Fund, Jian Tou Investment, Wanxiang Qianchao, and other top global state-backed funds and industrial capital firms. To date, this company—founded just three and a half years ago—has completed three consecutive rounds of large-scale financing within three months: a 1 billion yuan Pre-B round, a 1.5 billion yuan B1 round, and a 1 billion yuan B2 round, for a cumulative total of 3.5 billion yuan, with its valuation surpassing 10 billion yuan.

This is not just another round of funding. It is a collective vote of confidence by global capital in the technological path of ”world-model-driven physical AGI.”


Origin: A Tsinghua PhD’s ”First Principles”

On January 18, 2023, in Changping District, Beijing, a micro-enterprise with a registered capital of only about 2.12 million yuan was quietly established.

Founder Huang Guan—who holds a Ph.D. from Tsinghua University’s Department of Automation and gained hands-on experience at Microsoft Research Asia and Samsung China Research Institute—served as Head of Visual Perception Technology at Horizon Robotics, where he spearheaded the creation of WebFace260M, the world’s largest facial recognition dataset. Armed with a nearly stubborn conviction, he set the stage for Excellent Vision:

“World models are the next most important thing after language models.”

2023年初,ChatGPT刚点燃全球AI狂潮,所有人都在追逐大语言模型。黄冠却转身扎进了”物理世界的智能”——一个更难、更慢、但他认为更有价值的方向。

The team started with just 11 members. There was no fanfare or press conference—just a group of young people in the Intelligent Vision Laboratory of Tsinghua University’s Department of Automation who believed that ”AI will eventually move from the digital world to the physical world.”

His assessment is not without basis. As a technical leader who has repeatedly led teams to world championships in prestigious global AI competitions, Huang Guan is keenly aware that while digital AGI has already completely transformed the digital world, the intelligent transformation of the physical world is the true ”last mile.”


Growth: Four Funding Rounds in Three Years—From a ”Small and Micro” Startup to a ”10-Billion-Yuan Unicorn”

Jijia Vision's funding pace sets a record for acceleration in the hard tech sector:

timing classifier for laps, turns, rounds sum of money Key Investors
2025 Pre-A / Pre-A+ several hundred million dollars Guozhong Capital, CICC Capital, Guangzhou Industrial Investment
November 2025 Strategic investments Huawei Hubble Investment
December 2025 Round A2 200 million yuan Led by Dachen Capital, the project is being established in the Hengqin Guangdong-Macao Deep Cooperation Zone
March 2026 Pre-Series B Round One billion dollars. SMIC Juyuan, Shanghai Semiconductor Industry Investment, Linxin Capital, and others
April 2026 Round B1 1.5 billion yuan With a valuation exceeding 10 billion, it has become China’s first “world model” unicorn valued at over 10 billion.
June 2026 Round B2 One billion dollars. Lion City Capital, China-Belgium Fund, Jiantou Investment, Wanxiang Qianchao, etc.

Huawei's entry into the market with its Hubble technology is particularly crucial. In November 2025, Shenzhen Hubble Technology Investment Partnership, a subsidiary of Huawei, acquired a stake in the company, increasing its registered capital to approximately 2.117 million yuan. Huang Guan himself revealed: ”Huawei has ranked the World Model as the top technology trend among the ”Top Ten Technology Trends for the Future Intelligent World 2035,’ and this is also the underlying rationale for investing in Extreme Vision.”

This is not only an injection of capital, but also a strategic mutual recognition of technological approaches.

On June 20, 2026, Beijing Mayor Yin Yong visited Haidian District to investigate the development of the embodied intelligent robotics industry. He made a special effort to hear a briefing from Jijia Shijie on its technical roadmap and practical applications, and encouraged the company to ”keep pace with cutting-edge global trends in model technology, increase investment in the research and development of foundational models, and expand applications across diverse scenarios.”

From a small, 11-person startup to a unicorn valued at 10 billion; from a research paper in a Tsinghua University lab to robots up and running in a factory—Jijia Vision achieved all this in just three and a half years.


Core Technology: The World’s First ”Double Pyramid” System

The key reason why Excellent Vision was able to grow from a ”small and micro enterprise” into a unicorn valued at 10 billion in just two years lies in the fact that it has developed a systematic methodology—The World’s First ”Double Pyramid” System for Physical AGI.

极佳视界:从清华实验室到物理AGI"独角兽"的狂飙之路


The Fundamental Question: What Exactly Is Holding Back the Scaling of Physical General Intelligence?

Huang Guan's team's assessment was precise and unflinching: the two major bottlenecks—Lack of data, lack of algorithms.

  • Real-world data offers high accuracy but is costly and limited in scale;
  • Simulation data is scalable but suffers from the “sim-to-real gap”;
  • The mainstream VLA paradigm typically tokenizes visual and action data and feeds it into large language models, but this mechanism is inherently ill-suited for processing 3D spatial information, physical causal logic, and continuous action encoding.
The Data Pyramid (Five Layers)
Level data type Excellent View Layout
First Floor Internet Video Data High-Efficiency Reuse of YouTube, Panda-70M, and Others
Second Floor Real-Person Data U-01 Handheld Data Acquisition Hardware + Ego E-01 Data Acquisition Hardware
Third Floor World Model Simulator GigaWorld-0, a self-developed embodied world modeling platform
Fourth Floor Simulated Data High-fidelity synthetic data generated by World Model
Fifth Floor Real-world data Wheel-Arm Robot Shiguang S1 + Data Acquisition Hardware Maker M01

Expected to be achieved by the end of 2026A total of 1 million hoursTraining data that supports the model's strong generalization capabilities across different tasks, scenarios, and ontologies.

Algorithm Pyramid (Three Layers)
Level Core competencies
bottom (of a pile) Simulating the World: Understanding and Predicting the Physical World
Middle Management Action Alignment—Transforming Understanding into Action Strategies
Top Level Experience-Based Reinforcement—Self-Evolution in Real-World Environments


“The ”World Generation–Action” Dual-Model System

This is Excellent Vision’s true technological moat—World Action Modeltogether withWorld Generation ModelsA Spiral of Progress, Each Step Indispensable:


The World Action Model (Turning Understanding into Action)
::

mould Core competencies Key Achievements
GigaBrain-0 A Model-Driven, Embodied VLA Large Model RoboChallenge Global Hands-On Review51.671 TP4T mission success rate—the highest in the world, leading Physical Intelligence's π0.5 by nearly 10 percentage points
GigaBrain-0.5M* The World’s First Native Paradigm for ”World-Model-Driven Experiential Learning” in Physical Agents Success Rate for High-Difficulty, Long-Duration MissionsClose to 100%
GigaWorld-Policy A Global Action Model That Breaks the ”Speed-Performance-Efficiency” Impossible Triangle Inference speed and training efficiency have both increased tenfold, and the task success rate has risen by approximately 30 percentage points;RoboCasa365 Review: Number One in the World, the first global action model to top the charts


World Generation Models (Understanding, Simulating, and Generating the Physical World)
::

mould Core competencies Key Achievements
GigaWorld-0 A landmark study that is the first in the world to demonstrate that ”data generated by world models can effectively enhance the performance of real robots” Open-sourced in December 2025, GitHub received1.5k+ Stars
GigaWorld-1 The World's Leading Action-Conditional World Model (AC-WM) WorldArena Overall Score62.34, beating out Google, NVIDIA, Alibaba, and others,The first model on the list to break the 60-point mark
DriveDreamer The World's First Autonomous Driving World Model for the Real Physical World Invited for an NVIDIA Oral Presentation; One of the Most Influential Papers at ECCV 2024

It is worth noting that in April 2026, Alibaba’s embodied world model, ABot-PhysWorld, briefly surpassed GigaWorld-1 to take the top spot in WorldArena, but Jijia Shijie immediately responded by securing 1.5 billion in Series B1 funding and achieving a valuation exceeding 10 billion—the competition is far from over, but Jijia Shijie has firmly established itself in the top tier.


Product Portfolio: A Full-Stack, Closed-Loop Approach from the Brain to the Body

Excellent Vision is not a company that ”only does algorithms.” Its product philosophy is:World Model Platform × Embodied Foundation Model × General-Purpose Ontology, All three pieces are essential.

product lines representative product localization Progress
World Model Platform GigaWorld Infrastructure-level products driving two major areas: driving and embodied intelligence Has signed contracts with more than 30 automakers for mass production; DriveDreamer serves more than 30 automakers and autonomous driving companies
Embodied Foundational Model GigaBrain Series The Robot's ”Brain” GigaBrain-1 is set to be released in Q3 2026; GigaBrain-3 aims to achieve the ”GPT-3 moment” for physical AGI”
General Ontology (Industrial) Maker H01 Bimanual robotic arm, 20+ degrees of freedom, kilogram-class payload Mass production and delivery have begun, with the goal of reaching 1,000 units
General Ontology (Family) Shiguang S1 Wheel-arm configuration, designed specifically for home use We have received an order for 100 units and will begin full-scale operations in Q3.
automatic driving DriveDreamer A New Generation of Driving Simulators Powered by World Modeling Serving over 30 automakers and autonomous driving companies

The product philosophy behind the Shiguang S1 is worth noting—it doesn’t adopt a fully humanoid design, but rather a wheel-and-arm configuration. Huang Guan’s assessment is very pragmatic: ”The essential needs in a home setting are reliably carrying water, clearing the table, and handing snacks to children—not parkour in the living room.” The wheel-arm configuration offers comprehensive advantages in terms of stability, safety, range, and cost.


Implementation: Consumer-facing solutions enter households; business-facing solutions enter factories

The commercialization of "Excellent Vision" is proceeding along the following path:“Two-Track Approach”Route:

B2B: Industrial Applications Have Reached Scale

In April 2026, GigaWorld, in collaboration with FAW Mold and Alibaba Cloud to deploy the GigaWorld + GigaBrain + Maker H01 trio in a real FAW Molds factory. By addressing tasks such as pallet de-stacking, cross-area transport, dynamic obstacle avoidance, and precision operations, they reduced the adaptation cycle—which typically takes months with traditional automation solutions—to just a few weeks.

On June 3, 2026, Jijia Vision and Longsheng Weirui, a subsidiary of Longsheng Technology, entered into a strategic partnership to jointly deploy 1,000 general-purpose robots in Wuxi over the next three years, equipped with Jijia Vision’s World Model and Embodied Brain—marking the world’s first large-scale deployment of 1,000 general-purpose robots driven by a foundational physical intelligence model in an industrial setting.

Consumer Market: The Home Setting Is Poised for Growth

Shiguang S1 has secured 100 orders for use in real-life home settings in China. The Shiguang S2/S3 series is set to launch in Q3 2026, with the goal of achieving the ”ChatGPT moment” for physical AGI—enabling everyday skills to be widely applied in real-life home settings.


Outlook: How Far Away Is the ”ChatGPT Moment” for the Physical World?

Huang Guan presented a clear three-phase roadmap aligned with the evolutionary path of digital AGI:

stage Digital AGI Benchmarking Excellent View Corresponds To
Intelligent Emergence The GPT-3 Moment GigaBrain-3, trained on 10 million hours of video data
General Skills for All ChatGPT Moments Shiguang S2/S3 Series: A Breakthrough in Consumer-Grade Industrial Technology
Professional Skills Claude Code Moments Expert-Level Breakthroughs in Industrial Applications

Huang Guan said, ”The GPT-3 era saw the emergence of intelligent capabilities in models; the ChatGPT era brought productivity benefits to every ordinary person; and the Claude Code era saw digital intelligence models reach the level of experts in specialized fields. Physical AGI will also undergo similar stages in the future, with one key difference: physical AGI will directly impact the real physical world. It will not only improve information efficiency but also reshape the way we produce and live, and therefore its impact on the economy and society will be even more far-reaching.”


put at the end

The story of "Excellent Vision" is, at its core, a story about“First Principles”The story.

While everyone else was chasing large language models, Huang Guan set his sights on the next frontier—the physical world. He addressed the fundamental question of ”What underpins the scaling of physical AGI?” with his ”Dual Pyramid” framework, and solved the ”last mile” challenge of moving “from the lab to mass production” through a closed-loop, full-stack hardware and software solution.

From a small, 11-person startup to a unicorn valued at 10 billion; from a research paper in a Tsinghua University lab to robots operating in factories; from a record-breaking 3.5 billion in funding raised in just three months to the attention of the Mayor of Beijing—Jijia Vision has proven one thing:

True faith in technology doesn’t need fanfare to prove itself. It just needs time.

And time is on their side.

© Copyright notes

Related articles

No comments

none
No comments...