Robot Forward reported on 9 September that Beijing Physical-AI data-base company WuWen ZhiKe has closed a Series A of several hundred million yuan. The round was led by Hongtai Fund, with Hongshan Capital, Zhongdian Shurong and Lion City Capital following, and existing shareholders Bridge Capital, Lihe Financial, Lihe Zhongke and Chengkai Lingcheng oversubscribing; Shneng Capital acted as long-term financial adviser.
A world model times a data factory times a simulator
The proceeds will deepen Physical-AI infrastructure: world-model research, high-quality real physical-interaction data production through a SimReady asset system and world simulator, and the full Real2Sim2Real chain with standardised evaluation. The firm has built its Physical-AI data base called WuYin and a world simulator called WuQiong, tying real collection, automated production, world reconstruction, simulation, scaled evaluation and real-machine feedback into one loop.
On the real-world side, self-developed multimodal collection devices use a sensationless wild-collection paradigm to gather information at scale, and a world-understanding model turns that into an automated production pipeline. The firm says this lifts Raw Data to Model-Ready Data efficiency by 500 per cent to 1,000 per cent. Adding wild-collected data on top of open-source real-machine datasets raised basket-tidying success by about 20 per cent.
WuWen ZhiKe says it has accumulated tens of thousands of hours of high-quality real physical-interaction data, with monthly production above 20,000 hours, reaching into logistics warehouses, industrial manufacturing, home services and retail. On the simulation side, its SimReady asset build speed leapt from three days per asset to five minutes, roughly a 300-fold gain, and it claims the industry’s largest million-scale SimReady library.
Hardware to capture the data
Its bimodal world simulator lifts flexible-object simulation correlation above 90 per cent and keeps contact error under 1 per cent. On capture hardware, the WuYin Capture range includes CaptureSight, a first-person unit with a six-channel global-shutter camera, 270-degree field of view, 1,600 by 1,300 resolution and a six-axis IMU at 200 Hz; CaptureSuit, a full-body suit with plus-or-minus 16 g acceleration, plus-or-minus 2,000 degrees per second gyro, 0.01-degree resolution and 1,000 Hz sampling under 30 ms latency; and CaptureGlove, an inertial glove with matching specs and no magnetometer.
Commercially, WuWen ZhiKe reports embodied orders of several hundred million yuan, year-on-year order growth above 3,000 per cent, revenue growth above 1,000 per cent, top-client coverage above 80 per cent and a repurchase rate above 80 per cent. Customers named include ByteDance, WuJie Dongli and ZhangYu Dongli, with long-term ties to Xingdong Jiyuan, Tashi Zhihang, Lingxin Qiaoshou and Lingfang.
Founder and CEO Liu Shengxiang was Baidu’s autonomous-driving test lead and smart-car test lead, a top-twenty Apollo engineer who built Baidu’s test platform, AI labelling platform and largest autonomous-driving scenario library. CTO Yu Chunlei earned a doctorate at CNRS affiliate Heudiasyc in France and co-founded Chaoxing Weilai; chief architect Zhu Jianxin is a former Baidu architect. Prior rounds included a seed, a tens-of-millions angel and a nine-figure round backed by Zhejiang Chuangtou and Jiushi Intelligent.
More images from the source report:
Editor’s note: This is an adapted translation of the original Zhidx report. It has been trimmed and restructured for readability for an international business audience.
Translated and adapted from Zhidx (https://www.zhidx.com/p/592292.html).