POCO Robotics破壳机器人
LatestAngel (Apr 2026, tens of millions USD, $400M valuation)
$400M (post-Angel); next round expected at $700M (under negotiation)
POCO (Beijing) Intelligent Technology Co., Ltd.
A Tsinghua-lineage home robotics startup proposing the 'UAG architecture' (unifying perception-action-generation), with a pre-trained 32B-parameter embodied foundation model, aimed at letting robots manipulate unfamiliar objects the way humans do
数据截至2026-08-19
About
POCO Robotics was founded in Beijing in March 2026 as POCO (Beijing) Intelligent Technology Co., Ltd. by Xu Huazhe, assistant professor at Tsinghua University's Institute for Interdisciplinary Information and former Chief Scientist of 星海图, making it a 'star startup' in embodied AI. The company proposes the UAG architecture, unifying Understand, Act and Generate in a single architecture so robots can end-to-end close the loop of 'understand the environment - execute manipulation - generate new skills'; a pre-trained 32B-parameter embodied foundation model gives robots generalization to unfamiliar objects, and a world model is introduced to strengthen causal understanding of the physical world and task planning. Its body uses a wheeled base + dual arms to enter home scenarios, combining UMI (Universal Manipulation Interface) + video data collection to lower the data acquisition barrier; products remain in the R&D and demo stage (to be confirmed). From late March to April 2026 it completed an Angel round of tens of millions USD led by Yunqi Partners at a $400M valuation, with Xiaomi's strategic investment arm, 星海图, Baidu Ventures and 10+ other institutions participating; in August, negotiations for a new round began at an expected $700M valuation.
Technical Approach
Unifies perception, action and generation in the UAG architecture to end-to-end close the loop of 'understand the environment - execute manipulation - generate new skills'; a pre-trained 32B-parameter embodied foundation model addresses the limitation that traditional models 'can only manipulate what they have seen'; introduces a world model to strengthen physical causal understanding and task planning; enters home scenarios with a wheeled base + dual-arm body, using UMI + video data collection to lower the data acquisition barrier.
Commercialization（2）
- Early stage; no commercialization or order information disclosed yet (to be confirmed)
- As of August 2026, negotiations for a new round had begun at an expected valuation of $700M (source: DoNews, Mar 31, 2026)
Position & Strengths（4）
- Endorsed by Xu Huazhe of Tsinghua's IIIS with top academic influence — a 'star startup' in embodied AI
- Angel-round valuation of $400M (within 2 months of founding), an industry-leading pace reflecting exceptional capital recognition
- Differentiated technical route: UAG architecture + 32B pre-trained model, aimed directly at the 'generalized manipulation of unfamiliar objects' problem in home scenarios
- Jointly backed by multiple strategic investors (Xiaomi, 星海图, Baidu Ventures, etc.) with rich ecosystem synergies