KNOWIN

KNOWIN (Shenzhen KNOWIN Intelligence Co., Ltd.) was founded in August 2025 and is headquartered at A2406, Building 12, Shenzhen Bay Tech-Eco Park, No. 18 Keji South Road, Yuehai Street, Nanshan, Shenzhen. It is an embodied-AI company focused on consumer home scenarios under the banner of bringing intelligent robots into every home, treating the household as its biggest and only battlefield. Founder and CEO Li Yinchuan holds a joint PhD from Beijing Institute of Technology and Columbia University, joined Huawei at 25, became head of the generative-large-model direction at Huawei's Noah's Ark Lab at 27 and left at 30 to found the company, with R&D and deployment experience spanning embodied AI, foundation models and autonomous driving. He set out to build the world's most generalizable embodied foundation model and therefore chose consumer (To C) embodied AI over To B/To G, arguing that home robots are a key opportunity to turn large-model capability into a real product and that entering the home requires solving adaptability and cost at the same time. The company follows a fully self-developed integrated software-hardware route: its generative embodied foundation-model architecture GLOW (Generative Learning of World) handles perception and decision-making through four modules that close a loop of experience construction, world understanding, model training, action generation and execution feedback - KnowinGLOW (a unified multimodal autoregressive model folding the former Brain and Act capabilities into BrainSkill and ActSkill, jointly learning visual understanding, spatial reasoning, task planning and action generation), KnowinDream (synthetic embodied-experience generation, where a Life Engine compiles family roles, habits and home semantics into a Household Program, a Physics Engine yields physically feasible grounded trajectories and a Dream Engine renders photoreal experience, with an Experience Compiler aligning observations, actions, states and contacts), KnowinWorld (an interactive 3D world model for action-outcome rollouts and embodied alignment that reasons about consequences without generating full video frames) and KnowinAgent (Harness, managing task memory, tool calls, replanning and execution feedback). About 90% of its training data is synthetic, cutting overall data cost by 10-100x, and a single inference emits a complete action sequence to lower call frequency. The latest GLOW centers on one-shot teaching: a user's single demonstration is encoded into a reusable skill, and the robot tackles new tasks by reasoning over the live situation instead of retraining or updating parameters; real-robot demos validate object generalization on a watering task and triple generalization across skill, object and placement on unboxing and storage, because what it learns is task goals, object relations and task progress rather than one fixed trajectory. Benchmarks under the evaluation settings disclosed on its website: 62.2% average success across 18 RoboDojo simulation tasks, 40.0 percentage points above the GPT-6 (Robocurve) baseline; 86.7% average success on LIBERO-Pro, first among the single-policy models listed and 28.5 points above the GPT-6 Astra baseline; KnowinBrain-1.5 ranks No. 1 on the 2D embodied QA leaderboard of Embodied Arena with 65.62 (KnowinBrain took both the overall and core-dimension average first places in 2025); it also won both the Source-Limited and Open-Source tracks of the first EgoCross Challenge at the CVPR 2026 EgoVis Workshop. Its first consumer product, KNOWIN-X1, is a foldable dual-arm home robot with working and stowed configurations that folds to 40cm or less so it fits tabletops, floors, sofas and washing machines; the fully self-developed arms have 7 DoF each with a 6kg dual-arm payload and 23 DoF whole-body for bimanual coordination, on a ~50kg body with 10 hours of standby, a wheeled base, and vision-depth-LiDAR fusion plus multi-layer obstacle avoidance. The team had 150+ high-level R&D staff as of June 2026 (90% from Ivy League, Tsinghua and Peking, the top three Hong Kong universities, NUS and NTU, and C9 schools) and grew past 200 by September 2026 across algorithms, software and hardware engineering; more than two thirds of algorithm roles hold a PhD or above and half graduated from QS top-30 universities. Two former Huawei Genius Youth members recently joined: Chen Yingcong as multimodal-model scientist advisor (assistant professor and PhD supervisor in the AI thrust at HKUST-Guangzhou, CUHK PhD and MIT postdoc, 90+ papers in TPAMI, CVPR, ICCV and ECCV and an ICCV 2023 best-paper nomination) and Deng Shiyuan as partner heading Infra (CUHK CSE bachelor's and PhD, ICDT 2023 best-paper award, working on database theory, vector retrieval and heterogeneous computing), complementing the existing foundation-model, body and control teams. On September 29, 2026 it announced an Angel+++ round of several hundred million RMB led by a JD-related fund (an industrial investment fund under JD Group), followed by Loyal Valley Capital, Nanshan Strategic Emerging Industries Investment and Walden Investment - only 49 days after the single RMB 500M Angel++ round disclosed on August 10, 2026 (led by Matrix Partners China, with Redpoint China Ventures, SenseTime Guoxiang Capital, Walden Technology and the L2F Light Source Entrepreneurs Fund, and Lighthouse Capital as exclusive financial advisor). Five rounds in its first year put cumulative funding above RMB 1 billion. Proceeds go to three areas: expanding large-scale real-scenario training data to fix GLOW's generalization on unseen objects and unfamiliar floor plans; deepening body R&D to push KNOWIN-X1 through strict engineering validation and production readiness; and continuing to recruit top embodied-AI research talent to strengthen core-technology breakthroughs. KNOWIN-X1 is scheduled to enter launch and mass production in Q1 2027.

City

Shenzhen

Founded

2025

Ticker

Unlisted

Website

knowinai.com ↗

Team size

200 · 2026-09 官宣稿/36氪