RoboScience Secures Nearly 200 Million Yuan in Angel Round Funding

On July 30, 2025, it was reported that the startup RoboScience has completed nearly 200 million yuan in angel round financing. The round was led by JD Group, with participation from China Merchants Venture Capital and SenseTime Capital, and previous investor Zero One Capital also increased its investment. RoboScience, founded in December 2024, has rapidly developed since officially starting operations in March of this year, having previously completed several tens of millions in seed round financing. The company's founder and CEO, Tian Ye, is a graduate of the University of Science and Technology of China and Stanford University's AI Lab, having studied under Andrew Ng. He previously served as the technical head of machine learning platforms at Apple, leading the development of several core edge AI systems, including Apple Intelligence, and possesses extensive experience in large-scale edge AI deployment and ecosystem construction. Chief Scientist Shao Lin is currently an assistant professor at the National University of Singapore, and his proposed UniGrasp neural network architecture has become a benchmark method in the field of dexterous grasping, with related research winning the Best Paper Award at ICRA 2025 for robotics operation and motion. Co-founder Liu Penghai previously served as vice president of Ecovacs Robotics and has over 20 years of experience in robot product development and supply chain management, leading the mass production of more than 50 robotic products. Another co-founder, Wang Tao, was the fundraising head at SenseTime Capital and has over ten years of experience in technology investment and financing. Technologically, RoboScience uses a layered end-to-end model architecture with fast and slow brains. The fast brain layer is responsible for real-time response and dynamic adjustments (such as multi-joint coordinated control and real-time force feedback adaptation), while the slow brain layer focuses on deep logical analysis and long-term task planning, enabling high-precision complex operations with fully autonomous reasoning. To support model training, the company has developed its own simulation physics engine, defining "Object Trajectory" as a standard data format for embodied intelligence, achieving large-scale integration of simulated, video, and real data, effectively enhancing model generalization ability and reducing data collection costs. Based on this, they developed the Manipulation Foundation Model (MFM), which breaks through hardware and scene limitations, demonstrating significant improvements in generalization performance for dexterous grasping and providing a technical basis for any robot to perform any task with any object. The company's core technology modules include Cross Embodiment AI, a fast-slow brain robot learning framework, and a self-supervised training developed embodied operation system, which can efficiently adapt to diverse hardware platforms based on scene requirements. Currently, RoboScience is applying its integrated intelligent modules and complete systems across various fields such as industrial automation, logistics, consumer retail, and home services, aiming to promote the scalable deployment of intelligent robots.