Jijia Vision Secures Hundreds of Millions in Series A1 Strategic Financing Co-led by Huawei Hubble and Huakong Fund, Leading the Physical AI Towards the Ultimate Technical Route with the "World Model"
Breaking through three major technological bottlenecks, GigaVision launches a highly available world model system.
When AI capabilities begin to connect with the physical world, "embodied intelligence" becomes the implementation carrier.
Since this year, the "world model" has been rapidly gaining popularity in the field of embodiment: tech giants such as Google, OpenAI, Tesla, and NVIDIA have been intensively deploying in this area.
Many industry insiders believe that the world model will alleviate the bottlenecks of embodied intelligence in data scarcity and generalization difficulties. It is very likely to become the core technological trend in 2026 after VLA.
Against this background, GigaVision, a physical AI company focusing on the world model, completed three rounds of financing in the past two months and announced its latest progress.
According to "Intelligent Emergence", GigaVision recently completed a new round of hundreds of millions of yuan in Series A1 financing. This round of financing was jointly invested by Huawei Hubble and Huakong Fund. At the end of August, GigaVision announced the completion of two consecutive rounds of hundreds of millions of yuan in Pre-A & Pre-A+ financing.
The three rounds of financing in two months reflect the capital market's recognition of GigaVision's team strength, technological roadmap, and business progress. It also reflects the investors' judgment on the critical turning point of "general intelligence in the physical world" (physical AI).
Founded in 2023, GigaVision focuses on physical AI and is dedicated to "general intelligence in the physical world driven by the world model". Its products include the world model platform GigaWorld (for driving and embodiment), the embodied foundation model GigaBrain, and the general embodied ontology Maker, which are full-stack hardware and software products for physical AI.
In fact, in the view of GigaVision, the technological value of the world model has already been demonstrated at the current stage, and there is no need to wait until next year for its technological implementation: It not only improves the two problems of high-dimensional and high-quality data scarcity and the Sim2Real Gap of traditional simulators but also enhances the training effect of reinforcement learning.
What problems does the world model actually solve?
Put simply, the world model models the physical world and its operating laws in the digital world: It allows AI to build a simplified physical sandbox in its "mind" before taking action, predict what will happen in the next second, and choose actions accordingly, so as to make fewer mistakes and be more stable in unfamiliar environments.
Currently, several tech giants in Silicon Valley have entered the field of world models. NVIDIA launched the World Foundation Model - COSMOS to explore the application of world models in fields such as robotics and autonomous driving; Google DeepMind released Genie-3, focusing on high-precision modeling of complex dynamic environments; Tesla has also deeply integrated world model technology into its simulation system for the research and development of autonomous driving and robotics.
This fully demonstrates the global industry and academic community's emphasis on the world model.
Dr. Huang Guan, the founder and CEO of GigaVision, said: "Whether from the perspective of real business and technological needs or the consensus at the industrial and academic levels, the world model has become a key and popular direction for embodied intelligence. Huawei has also listed the world model as the top of the ten technological trends in the intelligent world in 2035, which is also the underlying logic for investing in GigaVision." In addition to investment, Huawei is also promoting strategic cooperation with GigaVision in multiple business lines.
According to the current progress of technology, Dr. Huang Guan predicts that the 'ChatGPT moment in the physical world' will arrive within two to three years.
Specifically, the world model will mainly solve the problem of generalization, while VLA is responsible for solving the complexity of tasks, and reinforcement learning solves the accuracy and reliability. When the three work together, they will lead physical AI to achieve a 95% success rate in 90% of the 100 common tasks.
Breaking through three major technological bottlenecks, GigaVision launches a highly available world model system
The core team of GigaVision is closely affiliated with the Intelligent Vision Laboratory of the Department of Automation at Tsinghua University. The team members include top researchers from well-known universities such as Tsinghua University and the Chinese Academy of Sciences, as well as senior executives from well-known companies such as Baidu, Microsoft, and Horizon Robotics. They have published more than 200 top AI papers, won dozens of global AI competition championships, and released several globally influential physical AI technological achievements.
Dr. Huang Guan, the founder & CEO of the company, is a doctor from the Department of Automation at Tsinghua University. He has research experience in well-known companies such as Microsoft, Samsung, and Horizon Robotics. He has led R & D teams of hundreds of people multiple times and has rich continuous entrepreneurship experience in the field of Physical AI. As a core executive, he has led or participated in financing of over 100 million yuan in total.
Based on in - depth insights into the industry, the team summarized the current bottlenecks of embodied intelligence into three major challenges:
1. The scarcity of high - quality data. Relying on real - machine data collection leads to high costs and low efficiency;
2. There is a Sim2Real Gap (simulation - reality gap) between simulation data and reality, making it difficult to directly apply;
3. The modeling errors of traditional simulators restrict the effect of reinforcement learning.
Zhu Zheng, the chief scientist of GigaVision, said that the company's embodied world model is a systematic solution to the above problems.
First, it can learn from a small amount of real data to build a unified model that understands the environment, tasks, and multi - modal features, and then generate a large amount of high - fidelity synthetic data to fill the data gap at extremely low cost.
Second, in response to the Sim2Real Gap of traditional simulators, the world model can integrate multi - modal feedback such as vision, touch, and force sense for continuous optimization. By finely modeling key errors, it significantly improves the authenticity and usability of the generated data.
In addition, the world model can also serve as a high - fidelity training environment for reinforcement learning, allowing strategies to iterate in highly realistic virtual scenarios, effectively avoiding the failure of strategy transfer, and thus greatly improving the performance of reinforcement learning in the real world.
Based on these technologies, GigaVision's GigaBrain - 0 has shown greater potential for performance improvement in comparisons.
Preliminary results show the performance advantages of GigaBrain - 0
Compared with other methods, it has more diverse sources of training data: it is more robust and has better generalization ability under changes in texture, lighting, and viewing angles;
It has a deeper architecture: deeper - level modeling is introduced in key sub - modules, resulting in more precise operation performance;
It has two versions, large and small: the small model can achieve about 90% of the effect of the large model and can perform real - time inference on the edge - side Orin.
Currently, GigaVision has reached in - depth cooperation with humanoid robot innovation centers, training fields, scientific research institutions, and cloud computing companies in many places, aiming to build a globally leading full - stack product including a virtual - real combined data factory and an embodied intelligence platform.
In the direction of implementing scenario solutions such as the world model and VLA large models, GigaVision has also reached in - depth cooperation with many global automotive industry giants, leading embodied ontology companies, and application scenario giants. All parties are jointly exploring the implementation of physical AI in multiple scenarios such as driving, industry, services, and households, accelerating the explosion of physical AI applications.
In the future, the company will continue to promote the R & D and iteration of physical AI intelligent models, accelerate the R & D of general embodied humanoid ontologies, and continue to create commercial applications in benchmark scenarios. Through the trinity of 'intelligence - ontology - scenario', it will accelerate the arrival of the 'ChatGPT moment in the physical world'.
Cover source | Visual China