Knowin Raises Another Angel Round, Total Funding Passes 1 Billion Yuan in First Year

Zhengxin Valley Capital, Nanshan Zhanxintou and Walden International joined the round, according to QbitAI, which published the announcement. The company's previous 500 million yuan round was disclosed on Aug. 10, 49 days before the latest one.

Knowin said the new money will go to three areas: expanding training data to improve GLOW's generalization across tasks, objects and home environments; deepening development of its robot body and advancing engineering validation and mass-production preparation for the KNOWIN-X1; and recruiting senior embodied-intelligence talent.

The company describes GLOW, its generative embodied model, as the robot's brain and the KNOWIN-X1 as its body. Its stated next step is to turn demonstrated abilities into a product that ordinary users can use over time, rather than showing what the model and hardware can do in a demo.

The updated GLOW is designed to learn a task from a single human demonstration. Once a user completes an operation, the system encodes it as reusable skill information; when a new task comes up, the model judges the next step from the current state of the scene instead of retraining or updating parameters, the company said.

In demos released by Knowin, a watering task showed object generalization, with the same task logic transferred to different vessels. A box-opening and tidying task showed generalization across skills, objects and positions: the robot could still organize subsequent actions after the way an item is handled changed, the item was swapped or its position moved. According to the company, the model learns task goals, object relations and task progress rather than a fixed motion trajectory, which means a robot's abilities are not fixed entirely by its factory feature list.

According to evaluation figures published on Knowin's website, GLOW achieved an average success rate of 62.2% across 18 simulated tasks in RoboDojo, 40 percentage points above the GPT-6 (Robocurve) baseline, and 86.7% on LIBERO-Pro, ranking first among the listed single-policy models. KnowinBrain-1.5 scored 65.62 to rank first on the 2D embodied question-answering leaderboard of Embodied Arena.

Knowin says it employs more than 200 R&D staff across algorithms, software and hardware engineering. More than two-thirds of the algorithm team hold doctorates, and half graduated from universities ranked in the QS top 30.

Chen Yingcong has joined as an advisory scientist for multimodal models. He is an assistant professor and doctoral supervisor in the artificial intelligence thrust at the Hong Kong University of Science and Technology (Guangzhou), holds a PhD from the Chinese University of Hong Kong, did postdoctoral research at MIT and received a top-tier offer under Huawei's Genius Youth program. He has published more than 90 papers in journals and conferences including TPAMI, CVPR, ICCV and ECCV, with research recognized by an ICCV 2023 best paper nomination and a first prize in natural science from the Chinese Society of Image and Graphics. Deng Shiyuan, who holds a bachelor's degree and a PhD in computer science and engineering from CUHK and was also selected for Huawei's Genius Youth program, joins as a partner and head of infrastructure; his research covers database theory, vector retrieval and heterogeneous computing, and his work received the ICDT 2023 best paper award.

Knowin says the KNOWIN-X1 will enter its launch and mass-production phase in the first quarter of 2027.