A young engineer who once took apart a game console to understand it grew a following by building robotic arms in his bedroom and posting the videos online. His nickname was "wild Iron Man." Today he runs a humanoid-robot company that rolled its 1,000th machine off a production line within two years of founding.
The jump from hobbyist to industrial builder is the story of Peng Zhihui (彭志辉), better known as Zhiyuanjun (稚晖君).
From DIY videos to Huawei's "Genius Youth"
Peng was born in 1993 in Ji'an, Jiangxi. He earned a bachelor's degree in 2015 and a master's in 2018, both from the University of Electronic Science and Technology of China (电子科技大学) — the bachelor's in life sciences and technology, the master's in information and communication engineering.
Under the handle Zhiyuanjun he began posting hardcore DIY engineering videos in 2017, eventually gathering more than 2.6 million followers on Bilibili. Fans dubbed him the "wild Iron Man" (野生钢铁侠) for projects that blended mechanics, electronics, and AI.
He first joined OPPO's research institute as an algorithm engineer in its AI lab. In 2020 he was hired through Huawei's "Genius Youth" (天才少年) program at its top annual package, reported at about RMB 2.01 million (≈ US$280,000 / HK$15.7 million), working on Ascend (昇腾) AI chips and AI algorithms. He left Huawei on 27 December 2022.
Founding AgiBot at speed
In February 2023, Peng co-founded AgiBot (智元机器人, legally Zhiyuan Robotics, based in Shanghai's Zhangjiang Science City), serving as co-founder and chief technology officer. The company moved fast: its first humanoid prototype, the Expedition A1 (远征A1), was unveiled in August 2023 — only six months after founding.
In August 2024, AgiBot launched five commercial humanoid (人形机器人) models at once — the Expedition A2, A2-W, A2-Max, and the Lingxi X1 and X1-W — spanning bipedal, wheeled, and interactive forms. On 6 January 2025, the company said its 1,000th general-purpose embodied robot rolled off the line, split between bipedal and wheeled units.
The open-data bet
What separates Peng's play from a typical hardware startup is the data strategy. AgiBot open-sourced a large real-robot dataset, "AgiBot World," with more than one million manipulation entries, and released GO-1, a generalist embodied foundation model built on vision-language-action architecture trained from its own fleet's demonstrations. The logic: robots learn from shared, scaled data the way large model (大模型) systems learned from scaled text.
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, Peng's willingness to open training data is less charity than a land-grab for the standard that the whole humanoid (人形机器人) ecosystem eventually adopts.
Why an inventor became a symbol
Peng's rise matters beyond one company. He is the visible face of a generation of Chinese engineers who grew up posting projects online, then jumped from big-tech labs into embodied AI just as the large model (大模型) made robots suddenly seem teachable. His credibility comes from having built things with his own hands, not from a management deck — a rare profile in a sector full of slideware.
That also raises the bar. Fans expect AgiBot to ship the future they watched him prototype; investors expect margins. The open-data strategy is his answer to both: turn personal craft into a shared, scalable pipeline.
Backing and scale
AgiBot attracted an unusually dense investor list — Tencent, BYD, HongShan (formerly Sequoia China), Hillhouse, and others — and by mid-2025 its valuation had crossed roughly US$2 billion. Reuters-reported sourcing later indicated the company was preparing a Hong Kong IPO with a targeted valuation in the tens of billions of Hong Kong dollars, though those plans were not yet public filings at the time of writing.
The company's product families — Yuanzheng (远征) for full-sized bipedal industrial use, Lingxi (灵犀) as a developer platform, and wheeled dual-arm systems — point at logistics, retail, hospitality, education, and early factory workflows, with shipments reported beyond China into Europe, North America, Japan, and Southeast Asia.
Why the open-data move is the real bet
Opening training data and a foundation model cuts against the instinct of most hardware startups, which guard any edge. Peng's logic is that humanoid usefulness is bottlenecked not by a single clever machine but by the scarcity of real-world manipulation data — the messy, varied examples robots need to generalize. By seeding a shared dataset and model, AgiBot is trying to set the lingua franca the ecosystem builds on, the way open frameworks tend to define a field. If outside labs and competitors adopt AgiBot World, the company's pipeline, not just its products, becomes the standard others train against.
Honest limitations
This is a robotics and embodied-AI profile, not a brain-computer interface (脑机接口) story; neural-link readers should look elsewhere. AgiBot's later unit-shipment and valuation figures come from secondary trackers and news aggregators and were not independently audited here; the company has not filed the detailed public financials a listing would require. Co-founder roles beyond Peng (notably ex-Huawei executive Deng Taihua as chairman) are reported but lightly detailed in English. Peng's own social-media fame can inflate coverage, so product claims should be weighed against delivered, deployed units rather than launch videos.
What readers can do now
- Track deployed units, not announcement clips. AgiBot's credibility rests on machines actually working in factories and warehouses, not on polished reveal reels.
- Watch the open-data moat. AgiBot World and GO-1 adoption by outside researchers is the real competitive signal — a standard, once adopted, is hard to dislodge.
- Compare the three leaders side by side. UBTECH, Unitree, and AgiBot now define China's humanoid (人形机器人) race; relative shipment, margin, and ecosystem pull matter more than any single founder's story.