Today, RoboScience announced the completion of… A nearly RMB 200 million angel-round financing was led by JD.com, with participation from China Merchants Capital and SenseTime Guoxiang Capital, while existing investor 01 Venture continued to invest. Mushi Capital served as the exclusive financial advisor for this round.

RoboScience was officially registered in late December 2024 and began formal operations in March of this year. Prior to this round of financing, it had already raised tens of millions of yuan in a seed funding round.

Founder and CEO Tian Ye He earned his bachelor’s degree from the School of Physics at the University of Science and Technology of China, and his master’s degree from Stanford University’s AI Lab, where he studied under Andrew Ng. He previously served as the technical lead for Apple’s on-device machine learning platform team.

During his seven years at Apple, he spearheaded the development and deployment of numerous groundbreaking AI technologies, including Apple Intelligence—the world’s first on-device–cloud‑coordinated large‑model inference system—along with the first on-device inference framework and a multi‑compute‑unit collaborative computing architecture, the first compiler‑fusion system, and the first on-device dynamic neural network. These innovations have helped build Apple’s cross‑platform AI ecosystem, enabling thousands of apps to launch and serving over one billion users across two billion devices, while amassing extensive expertise in large‑scale on-device AI deployment and ecosystem development.

Founder and Chief Scientist Shao Lin He is an assistant professor in the Department of Computer Science at the National University of Singapore. He earned his bachelor’s degree from the Department of Earth Sciences at Nanjing University and his Ph.D. from the Stanford University AI Lab, where he studied under Jeannette Bohg and co-supervised by Leonidas J. Guibas.

Shao Lin’s deep neural network architecture, UniGrasp, has become a benchmark approach for data-driven dexterous hand grasping. His cross‑entity dexterous grasping method, D(R,O), was awarded the Best Paper in Robotics Manipulation and Locomotion at ICRA 2025—the first time in the past five years that an Asian institution has won this honor as the primary affiliation—and he initiated and completed the Concept2Robot project as early as 2020, exploring the integration of natural language and video for learning robotic manipulation tasks, one of the pioneering efforts in the VLA domain. Furthermore, his proposed SAM‑RL method was nominated for the RSS 2023 Best Systems Paper Award, among other accolades.

Co-founder Liu Penghai He previously served as Vice President of the ECOVACS Group, General Manager of Kaihang Motor (ECOVACS Motor Co., Ltd.), and a core member of both the Company’s Strategy Committee and the Robotics Product Committee. With over 20 years of experience in new product development and launch, he built ECOVACS’ product development process and integrated supply chain management system from the ground up. He has led teams of more than 3,000 employees, oversaw the mass production of over 50 robotic products, and helped grow the company’s annual robotics revenue from RMB 800 million to RMB 8 billion. Drawing on 16 years of management expertise at Fortune Global 500 companies including Philips, General Electric (GE), and TTi, he possesses extensive experience in large-scale product manufacturing and commercialization.

Co-founder Wang Tao A graduate of the Department of Statistics at the University of Science and Technology of China, he previously served as Head of Fundraising at SenseTime Guoxiang Capital, where he spearheaded the fundraising and deployment of multi-billion‑yuan industry funds. With nearly a decade of experience in private equity and investment banking, he has led investments in dozens of companies across AI, integrated circuits, and cutting-edge technology sectors, amassing extensive expertise in corporate financing and investment, transaction structuring, M&A, and IPO projects.

On the technology front, RoboScience has adopted a hierarchical end-to-end model of fast and slow brains since its inception.

The fast brain handles real-time responses and dynamic adjustments—such as multi-joint coordinated control, real-time force‑feedback adaptation, and low-level object manipulation skills—acting as the “cerebellum” to ensure operational precision. Meanwhile, the slow brain focuses on deep logical reasoning and long‑term task planning—for example, deciphering complex instruction manuals to assemble furniture or analyzing human‑demonstrated tie‑tying steps to learn the skill—serving as the “brain” that orchestrates overarching tasks and enables fully autonomous reasoning, zero remote intervention, high precision, high complexity, and robust, interference‑resilient long‑range embodied manipulation.

Underpinning the deployment of this model is RoboScience’s fully in-house‑developed physics‑based simulation engine. Grounded in first principles, it defines “Object Trajectory” (the evolution of an object’s state) as the standard data format for embodied intelligence, enabling large‑scale acquisition and utilization of simulated, video, and real‑world data—covering a wide array of everyday objects, tasks, and scenarios. By fusing cross‑modal data, the model enhances its generalization capabilities; meanwhile, complementary validation using both simulated and real data ensures high data quality while significantly reducing collection costs, thereby providing efficient, sustainable support for continuous technological advancement.

Building on this foundation, RoboScience has independently developed the Manipulation Foundation Model. This large-scale embodied manipulation model serves as a bridge between multimodal large models and the physical world, achieving generalization across three key dimensions: guiding any robot, manipulating any object, and completing any task. Taking grasping as an example, compared with existing approaches that are limited to specific objects and robotic arms, this model decouples from hardware and delivers significant improvements in success rate, pose diversity, and computational speed, offering a new paradigm for dexterous grasping.

Leveraging highly generalized technological capabilities, the company’s products can efficiently adapt to hardware configurations tailored to specific application scenarios, enabling flexible expansion across diverse deployment contexts. At the heart of this offering is Cross‑Embodiment AI, a core technical module that, when combined with an embodied operating system developed on a fast–slow brain–based robotic learning framework and self‑supervised training, forms the product’s intelligent core.

RoboScience, guided by the vision of “bringing intelligent robots into every home and empowering countless industries,” is integrating its hardware‑software‑integrated intelligent modules and complete robotic systems across a wide range of sectors—including manufacturing, logistics, consumer retail, and the home—delivering fast, safe, smart, and user‑friendly solutions in diverse application scenarios.