PsiBot一年半融资20亿，押注数据手套

做的是什么生意

Providing an embodied intelligence 'brain' for logistics/retail scenarios: end-to-end VLA models plus self-developed data collection gloves and dexterous hands; after customer deployment, on-site data flows back to feed the models.

启动资金：Angel round + Pre-A round cumulative about 2 billion yuan RMB (20亿元人民币) (first disclosed in March 2026); seed round invested by Hillhouse Venture Capital (高瓴创投), BlueRun Ventures (蓝驰创投), and AgiBot (智元机器人).

起因

At the end of 2024, Wang Qibin (王启斌) (who had been responsible for Yunji Technology (云迹科技) hotel delivery robots and JD Logistics (京东物流) robot R&D) and post-2000s Chen Yuanpei (陈源培) (Peking University (北大) Yang Yaodong (杨耀东) team, Stanford (斯坦福) Fei-Fei Li (李飞飞) lab) and others founded PsiBot (灵初智能). The seed round received investment from Hillhouse Venture Capital (高瓴创投), BlueRun Ventures (蓝驰创投), and AgiBot (智元机器人), and the company co-established an embodied dexterous manipulation joint laboratory with Peking University (北大). The team judged: simulation data has a sim-to-real gap, real-robot teleoperation is too expensive, and what embodied intelligence truly lacks is low-cost, large-scale real interaction data.

经过

In the second half of 2025, the company pivoted internally: it stopped pure demonstration-type demos and fully shifted to real data collection and segmented scenario delivery; it successively completed the angel round (China Development Fund (国开金融), Guozhong Capital (国中资本), CCTV Converged Media (央视融媒体) and other 'national team' (国家队) and industrial capital) and the Pre-A round (led by Xuhui Capital (徐汇资本)), cumulatively about 2 billion yuan (20亿), with valuation rising about 6-7 times in one year. It self-developed the Psi-SynEngine multimodal data glove; collection cost is about 1/10 of real-robot teleoperation; in customer warehouses' clothing package supply scenario it achieved generalization on more than 1,000 items of clothing and a maximum of 800 UPH, entering the 'production accompaniment' stage.

结果

Still expanding: in 2026, it plans to scale shipment/delivery for three logistics scenarios—clothing package supply, box-in inspection, and sorting wall—build the largest dexterous hand dataset in China, and then extrapolate to more scenarios.

可借鉴之处

技术路线未定时，押注成本和数据反馈：PsiBot把数据成本降到1/10，让迭代复利起来。