Led by Chinese people, a 20 - member team with an average age of less than 28 has secured the largest seed - round financing for embodied intelligence in Silicon Valley.

Recently, Genesis AI, a U.S. startup in embodied intelligence, announced the completion of a $105 million (approximately 752 million RMB) seed - round financing. The round was co - led by top U.S. venture capital firms Khosla Ventures and Eclipse, with participation from former Google Chairman Eric Schmidt, European tech tycoon Xavier Niel, the French central bank BPI, and Sequoia China.

This is the largest seed - round financing ever completed by a Chinese - led team in Silicon Valley and also the largest seed - round financing in the embodied intelligence track in Silicon Valley to date, surpassing the previous record set by the star startup Physical Intelligence in this track.

Regarding this investment, Kanu Gulati, a partner at Khosla Ventures, said that the firm is betting on whether someone can develop a general robot foundation model that can handle multiple tasks. Among all the teams the firm has contacted, Genesis AI's uniqueness lies in its development path in the field of robot foundation models.

Genesis AI was founded in December last year. Its development team consists of about 20 people and is committed to building a general model that enables robots to automatically perform various repetitive tasks, from laboratory work to household services.

01. Led by a Chinese - led team with an average age of under 28

The founding team is composed of senior academic and industry talents from Mistral AI, NVIDIA, Google, Apple, CMU, MIT, Stanford, Columbia University, and the University of Maryland, with deep accumulations in fields such as physical simulation, graphics, and robotics.

Zhou Xian, the founder and CEO, graduated from Nanyang Technological University with a bachelor's degree and from Carnegie Mellon University with a doctorate. During his doctoral studies, his research areas covered world models, imitation learning, and reinforcement learning. He proposed a new paradigm for generative simulation and initiated and led the Genesis open - source project.

Théophile Gervet, the co - founder, graduated from Carnegie Mellon University with a doctorate in 2023. He is not only one of the founding members of Skild AI but also the former head of the multimodal model at Mistral AI.

Zhou Xian on the left, Théophile Gervet on the right

Xu Zhenjia graduated from Columbia University with a doctorate in 2024. He studied under Song Shuran, the recipient of the Sloan Research Fellowship. He is the co - first author of the previous Stanford dish - washing robot UMI project and the robot network architecture Diffusion Policy. He is also the person in charge of the real - machine system of NVIDIA's GEAR team and has led the robot data collection and real - machine system work of the GR00T project.

Xu Zhenjia

Wang Zunxuan graduated from MIT CSAIL with a doctorate. He studied under Daniela Rus, a well - known figure in the robotics field. His research areas span core technical directions of embodied intelligence such as data, the brain, and the body. He was previously responsible for the research and development of the foundation model architecture at the large - model company Liquid AI.

Wang Zunxuan

Qiao Yiling graduated from the University of Chinese Academy of Sciences with a bachelor's degree and from the University of Maryland with a doctorate. He studied under Ming C. Lin, a leading figure in graphics, and Dinesh Manocha, a leading figure in robotics. He is the recipient of the Meta Scholarship and has made several groundbreaking achievements in graphics and differentiable physical simulation.

Qiao Yiling

Li Minchen graduated from Zhejiang University with a bachelor's degree and from the University of Pennsylvania with a doctorate. He is an assistant professor in the graphics direction at CMU and the inventor of the IPC algorithm, the most influential algorithm for solving complex contact models in the graphics field.

Li Minchen

Song Yunlong graduated from the University of Zurich with a doctorate. He studied under Davide Scaramuzza, a leading figure in robotics.

Song Yunlong

In addition to the above academic talents, the early founding team members also include several engineering experts: the head of the multimodal model at Apple Intelligence, Google engineering director Rachid El Guerrab, NVIDIA GVDB architect Rama Hoetzlein, and Antoine d’Andigne, the head of the multimodal model at H Company.

02. Has released a physical world engine, becoming the largest open - source project in embodied intelligence

At the end of last year, an open - source generative physical engine project called Genesis attracted much attention. The project was led by Zhou Xian and jointly carried out by more than 10 research institutions, including Carnegie Mellon University, Tsinghua University, Peking University, MIT, and Stanford.

Several researchers who participated in the project have joined Genesis AI, forming a development team of about 20 people, specializing in the fields of robotics, machine learning, and graphics.

Genesis has received 25.4k stars on GitHub and has grown into the largest open - source project in embodied intelligence.

Just as large language models require a massive text dataset for training, embodied intelligence models require data from the physical world for training. However, obtaining data from the physical world is an extremely expensive and time - consuming process.

Therefore, Genesis AI has chosen the synthetic data route to generate data through a physical generation engine.

The Genesis physical generation engine can generate a 4D dynamic world, simulate a wide range of materials and physical phenomena, and is specifically designed for general robots, embodied AI, and physical AI applications.

It can provide "a simulation speed about 430,000 times faster than the real world." Even in a single - card training environment with an RTX4090, it only takes 26 seconds to train a motion strategy that can be transferred to a real robot.

Currently, Genesis AI is developing its synthetic data and building a foundation model in two offices in Silicon Valley and Paris. It is expected that by the end of this year, the company will release an embodied intelligence model to the robotics community.

03. Conclusion: The embodied intelligence track is heating up, accelerating the evolution towards general intelligent agents

Genesis AI's record - breaking seed - round financing also means that the industry's strategic bets on general models for multiple tasks have entered a white - hot stage.

With the decreasing data cost and the improving generalization ability of multimodal large models, embodied intelligence is accelerating its evolution from a dedicated tool to a general intelligent agent. This technological revolution led by a Chinese - led team may redefine the interaction boundary between humans and the physical world.

This article is from the WeChat official account "Robot Outlook". Author: Xu Lisi, Editor: Mo Ying. Republished by 36Kr with permission.