Guanglun Intelligence Completes Tens of Millions of Yuan in Pre-A+ Round Financing, Accelerating Spatial Intelligence with Synthetic Data | 36Kr Exclusive

Text｜Xu Caiyu

Editor｜Li Qin

Recently, Guanglun Intelligence (Beijing) Technology Co., Ltd. (hereinafter referred to as "Guanglun Intelligence") has completed a Pre-A+ round of financing of tens of millions of RMB, led by the Beijing Artificial Intelligence Industry Investment Fund, with Matrix Partners China, a shareholder, continuing to follow the investment. This round of financing will be used for the research and development and upgrade of autonomous driving and embodied intelligence on the end-to-end technical route, further business expansion, and acceleration of global business expansion.

Established in 2023, Guanglun Intelligence is committed to providing enterprises with data and data-centric end-to-end solutions. This solution can amplify the utility of human demonstrations through synthetic data, while using closed-loop simulation to accelerate Self-play RL (Self-play Reinforcement Learning), thereby ultimately achieving Spatial Intelligence.

Since its establishment, Guanglun Intelligence has completed multiple rounds of financing, with investors including SEE Fund Infinity Fund, MiraclePlus, Chentao Capital, and other well-known venture capital institutions.

The company's founder and CEO, Dr. Xie Chen, previously served as the head of autonomous driving simulation at NVIDIA, Cruise, and NIO, building the synthetic data system of the above-mentioned enterprises from scratch. He is an international pioneer in integrating generative AI into simulation.

The chief scientist, Zhao Hao, is an assistant professor at the Institute for AI Industry Research, Tsinghua University. He once led the research and development of the world's first open-source modular realistic autonomous driving simulator, MARS. The company has formed a cross-disciplinary team with international top-level generative AI, simulation, synthetic data research and development, and implementation capabilities.

In terms of research and development achievements, Guanglun Intelligence provides 3D hybrid rendering, physically realistic, highly interactive synthetic data, and data-centric end-to-end solutions based on the core technology stack of "Real2Sim2Real + Realism Validation".

"Simulation technology is the key path to realizing the implementation of end-to-end autonomous driving and embodied intelligence algorithms. Different from the traditional simulation for the purpose of service evaluation, Guanglun Intelligence is based on training-centered simulation, deeply combines generative AI and the evaluation ability of the authenticity and utility of synthetic data to efficiently generate high-quality synthetic data, quickly support the needs of high-level mass production of autonomous driving, and has served the world's top embodied intelligence companies." said Xie Chen, CEO of Guanglun Intelligence.

Since the birth of Tesla's FSD V12 autonomous driving system, a new trend of the end-to-end technical route of autonomous driving has emerged at home and abroad.

However, due to the change of the technical architecture and the demand for adaptability to the complex environment, data has become the core to improve the effect of the end-to-end algorithm. Correspondingly, data capability has naturally become the main battlefield for the competition of the new generation of autonomous driving algorithms.

Among them, synthetic data can amplify the value of real data, effectively solve the pain points such as the difficulty of data collection, the high cost of annotation, and the slow iteration speed, and has become an efficient solution to the "data hunger" dilemma under the end-to-end architecture of companies such as Tesla, OpenAI, and NVIDIA.

Guanglun Intelligence has layouts in both the data Pre-train and Post-train stages, which can not only support model training based on imitation learning but also build a new training and evaluation paradigm based on Self-play RL (Self-play Reinforcement Learning) through closed-loop simulation.

In addition to autonomous driving, the data shortage in the field of embodied intelligence such as robots is more obvious. The real data used by Google comes from 13 robots, and it takes 17 months to collect. The cost and efficiency of data collection have become the bottleneck for the model's ability to climb.

The synthetic data provided by Guanglun Intelligence is based on a realistic physical model and a strong generalization ability. It, together with real data and Internet data, constitutes the data pyramid for model training to serve the evolution of embodied intelligence algorithms.

In September of this year, Guanglun Intelligence's synthetic data solution won the Global Innovation Award from Bosch. In terms of business progress, many breakthroughs have also been made.

Among them, in the field of autonomous driving, Guanglun Intelligence has signed contracts and delivered services with many domestic and foreign leading OEMs and Tier 1 suppliers, providing data services for several Chinese independent brands to go global, providing Corner Case (Long-tail Scenario) data for Tier 1 suppliers and autonomous driving companies, and cooperating with many leading OEMs to accelerate the mass production and implementation of end-to-end algorithms.

In terms of embodied intelligence, Guanglun Intelligence has reached a cooperation with the US-based embodied intelligence company Figure AI.

Investors Say

Beijing Artificial Intelligence Industry Investment Fund: Data scarcity is an important bottleneck for AGI. Guanglun Intelligence's unique technology and synthetic data solution have great potential to solve the data bottleneck. The Guanglun team has comprehensive technical accumulation and implementation experience. In more than one year since its establishment, it has won orders from domestic and foreign leading OEMs, Tier 1, and embodied intelligence companies, and its commercialization has entered a stage of rapid development. We are optimistic about the value of Guanglun as a data infrastructure company in the domestic AI infrastructure layer, and we believe that it can help the rapid development of AI through data.

Wang Huadong, Managing Partner of Matrix Partners China: Guanglun Intelligence is actively laying out the international market, and its embodied intelligence and autonomous driving businesses are expanding rapidly overseas and commercially. Compared with autonomous driving, the Scaling Law of embodied intelligence model training is more complex, and the VLA model requires a large amount of training data. As the next generation of Scale AI, Guanglun Intelligence's data delivery professionalism and engineering capabilities have a unique competitive advantage in the global market. We believe that Guanglun Intelligence is rapidly becoming an international top enterprise in the field of synthetic data and occupying a global leading position.