AI Robotics Startup TranscEngram Secures Major Seed Funding

A new contender in the advanced robotics sector, TranscEngram (also known as Yi Sheng Technology), has successfully closed a substantial angel funding round, amassing hundreds of millions of yuan. This significant capital injection is set to accelerate the development of its novel memory-based learning systems designed for embodied artificial intelligence. The company, founded by the esteemed Professor Yi Ma from the University of Hong Kong, aims to revolutionize how robots perceive, learn, and interact with the physical world, moving beyond the limitations of current large language models.

Professor Yi Ma, a distinguished figure in computer vision and recipient of the prestigious David Marr Prize, co-founded TranscEngram in September 2023 alongside Professors Gao Shuanghua and Yang Yanchao. Their vision centers on creating interpretable autonomous intelligence through a unified 'brain+cerebellum' architecture. This approach seeks to imbue robots with a more biological form of cognition, integrating perception, prediction, and interaction into a cohesive operational loop. The company's technological foundation is built upon memory-based world models, a departure from static, knowledge-based AI systems that often struggle with real-world validation and exhibit 'hallucinations'.

The impressive funding round saw participation from a diverse group of investors, underscoring broad confidence in TranscEngram's innovative strategy. Key backers include China Biologic Products (a subsidiary of Zheng Da), Pudong Innovation Fund, Zhang Jiang Technology Investment, Zhang Jiang High-tech, Hongxin Electronics, Yun Hui Capital, Volcan Capital, and Gold Anchor Capital. These funds will be strategically deployed to advance research and development in areas such as explainable embodied control foundation models, physically grounded world models, and extensive multi-modal datasets for humanoid interaction. Expansion of the R&D team and the establishment of commercialization hubs in Shenzhen's Qianhai and Shanghai's Zhang Jiang are also priorities.

TranscEngram's core innovation lies in its dual-memory system, designed to mimic biological learning. The 'Visual Memory' component functions akin to a brain, processing environmental data to build spatial and geometric understanding for complex reasoning. Complementing this is the 'Motor Memory,' which acts like a cerebellum, refining motor control through interaction to achieve stable and high-frequency execution of tasks. This architecture promises significant performance gains, with early indications showing over a threefold improvement in multi-task learning compared to existing Vision-Language-Action (VLA) models. The system's ability to handle diverse tasks like coffee-making and laundry folding with high success rates, coupled with its capacity for zero-shot generalization and skill transfer across different robotic embodiments, highlights its potential to overcome current limitations in robotic dexterity and adaptability.

The company has structured its offerings into a four-pillar product suite: EngramTeleOp for intelligent teleoperation data collection with ultra-low latency; EngramEgo for capturing real-world humanoid motion data from wearable devices; EngramControl for distilling demonstrations into reusable motion memories enabling rapid learning; and EngramNav for environmental memory and navigation. These integrated tools are designed to streamline the development and deployment of sophisticated robotic capabilities. TranscEngram is initially targeting high-end hotel services and complex manufacturing assembly, particularly in sectors like aerospace, where precision and adaptability are paramount.