Who Is Peng Zhihui? The Bilibili 'Iron Man' Behind AgiBot

AgiBot shipped more humanoid robots than any company on Earth in 2025 — and the engineer who became its public face didn't make his name in a corporate lab. He made it on Bilibili, China's answer to YouTube, where millions watched him build a self-driving bicycle in his spare time. His name is Peng Zhihui (彭志辉), though most of China knows him by his handle, Zhihui Jun (稚晖君), and his nickname: the Wild Iron Man (野生钢铁侠).

His arc is one of the best stories in robotics right now — a self-made maker-celebrity who took a two-million-yuan salary at Huawei, walked away from it to start a company, and by the age of 33 was both the chief technologist of a multi-billion-dollar humanoid firm and the chairman of a listed company. So let's get to know him: where he came from, the viral builds that made him famous, and why AgiBot is the company everyone now mentions in the same breath as Unitree.

First, the short answer: who is Zhihui Jun?

Peng Zhihui is the co-founder and chief technology officer of AgiBot (known in Chinese as Zhiyuan Robotics, 智元机器人), and the most recognizable face of a company he helped start in 2023. The important nuance up front is that he is not AgiBot's CEO: that's Deng Taihua (邓泰华), a former Huawei vice-president who ran Huawei's computing business and serves as AgiBot's chairman and CEO. Peng is the company's President and CTO — the engineer-celebrity who gives AgiBot its public personality, much the way Wang Xingxing is the face of Unitree.

Unlike a founder such as Figure's Brett Adcock, who came to robots as an outsider businessman, Peng is a hardware-and-software engineer to his core — the kind Chinese fans half-jokingly call a "stack-overflow engineer" for the sheer breadth of what he can build himself. To understand AgiBot, you first have to understand the maker.

The Bilibili "Iron Man"

Peng started posting under the name Zhihui Jun on Bilibili in 2017, and his videos were not casual unboxings — they were hardcore, end-to-end engineering builds, documented with a fast, self-deprecating wit. By the end of 2022 he had more than two million followers; today he has more than 2.8 million, and in January 2022 he was named one of Bilibili's "Top-100 creators of 2021." Chinese tech media dubbed him the "Wild Iron Man" for builds that looked like they belonged in a Marvel film.

The project that broke him out internationally was XUAN — an autonomous, self-balancing bicycle he built largely on weekends over about four months and released in June 2021. With no rider, the bike stays upright using a spinning flywheel, then steers, plans paths, follows objects and avoids obstacles on its own — running its models on a Huawei Ascend 310 AI chip wired to a lidar and a depth camera, all for under 10,000 yuan in parts, and open-sourced on GitHub. It was picked up by Slashdot, autoevolution and DeepLearning.AI, and even James Gosling, the creator of Java, called it "amazingly well done."

He kept going. "Dummy" was a miniature, self-built robotic arm he demonstrated delicately suturing a grape. ElectronBot was an open-source USB desktop robot with a round LCD face that picked up roughly 2,900 GitHub stars in five days. There was a coin-sized Linux computer, a modular mechanical keyboard, an NFC business card with a screen. Tellingly, Peng rejects the "genius" label he's saddled with: he describes his work as simply "interest-driven" (兴趣使然), plus, in his words, "effort and luck" and good timing.

Before he ever sold a robot, Peng Zhihui had already proven the one thing that matters most in this field: that a single engineer, working alone on weekends, could make a machine perceive the world and act in it. AgiBot is that same instinct, scaled up with a few hundred million dollars.

Huawei's "Genius Youth"

Peng's path ran through one of China's most prestigious tech jobs. After a master's at UESTC and two years as an AI algorithm engineer at OPPO, he joined Huawei in 2020 through its elite "Genius Youth" (天才少年) program — a scheme Huawei launched after the 2019 US sanctions to lure top talent with eye-watering pay. Peng landed in the top salary band, reported at up to about 2.01 million yuan a year (roughly US$287,000), and worked as an "Ascend AI Edge Computing Specialist," doing full-stack R&D on Huawei's Ascend (昇腾) chips.

It's worth clearing up a common myth here: the viral bicycle and robotic arm were his personal hobby projects, not his Huawei assignment. His day job was chips and edge computing — and he was, by all accounts, very good at it, reportedly earning straight-A performance reviews and internal awards before he left. Then, at the end of December 2022, he announced on Bilibili that he was quitting one of the best-paid engineering jobs in China to "start a new cause… to do something more challenging." A few weeks later, that cause had a name.

Founding AgiBot (2023)

AgiBot was founded in February 2023 in Shanghai, built around a thesis that has since become the consensus of the entire industry: that the combination of cheap, capable hardware and modern AI "brains" makes general-purpose humanoid robots one of the largest markets imaginable. Peng built the machine intelligence; the company was anchored by his fellow ex-Huawei heavyweight Deng Taihua and an original team that also included Shanghai Jiao Tong University professor Yan Weixin on the hardware side.

The robots came fast, organized into three families: the full-size Yuanzheng ("Expedition", 远征) humanoids, the lighter interactive Lingxi (灵犀) bipeds — one of which, the Lingxi X1, AgiBot fully open-sourced in late 2024 — and the industrial-grade Genie line. But the piece that signals Peng's real ambition is the software. In March 2025 AgiBot released GO-1 (Genie Operator-1), a general-purpose "embodied" AI model built on a vision-language-action architecture the company calls ViLLA, trained on more than a million real robot demonstrations. It also open-sourced AgiBot World, a massive manipulation dataset gathered by a fleet of 100 robots — a deliberate bid to become a platform, not just a hardware vendor. It's the same "own the robot's brain" logic we traced across the industry in the AI systems behind the humanoid race.

From demo to factory: 10,000 robots

Plenty of startups can stage a slick demo. AgiBot's harder achievement is manufacturing — the test we keep coming back to in how you actually judge a humanoid robot. The numbers tell the story of an exponential ramp. AgiBot built its 1,000th general-purpose robot in January 2025 (it took nearly two years to get there), its 5,000th in December 2025, and its 10,000th — an Expedition A3 — on March 28, 2026. It went from 5,000 to 10,000 in roughly three months, and claims to be the first company in the world to mass-produce humanoids at that scale.

The independent scorecard backs the bravado: research firm Omdia ranked AgiBot No. 1 worldwide in humanoid-robot shipments for 2025, with more than 5,100 units and roughly a 39% global share. For a flourish, an AgiBot "Expedition A2" set a Guinness World Record in November 2025 for the longest continuous walk by a humanoid — 106.286 kilometres, from Suzhou to Shanghai's Bund, swapping batteries without ever powering off. The company says its robots are now working in logistics, retail, hospitality, education and factories across multiple continents, though it hasn't named specific customers.

The money — and the listed-company twist

Capital has poured in exactly as our piece on the coming robot boom predicted it would. In March 2025, Tencent led a funding round — its first-ever investment in embodied AI — that valued AgiBot at about 15 billion yuan (~US$2.1 billion). The cap table reads like a who's-who of Chinese tech and capital: JD.com, BYD, SAIC, Hillhouse, HongShan (Sequoia China) and LG Electronics among them, with more than 2.7 billion yuan raised in about a year and a half.

Then came the move that made Peng a national headline. In July 2025, AgiBot announced it would take control of a STAR Market-listed company, Swancor New Materials (上纬新材) — the first time a Chinese embodied-AI startup had seized control of a publicly traded firm. The market went berserk: Swancor's shares rose more than 1,300% in the months after the announcement. And on November 25, 2025, the listed company's new board elected its chairman: Peng Zhihui, then 33 years old. AgiBot insists the deal is not a backdoor listing — it's pursuing a separate Hong Kong IPO, reportedly targeting a valuation of HK$40–50 billion — but for a maker who was filming bicycle videos in his bedroom a few years earlier, being handed the gavel of a listed company is a remarkable plot twist.

AgiBot vs Unitree: China's "twin stars"

You cannot tell Peng's story without the rivalry that defines it. Chinese and international media frame AgiBot and Unitree as the "twin stars" of China's humanoid industry — and their founders as the sector's poster children. Together, forecasters expect the two companies to account for close to 80% of global humanoid shipments in 2026. It's a genuine duopoly, and the contrast between the two men is irresistible.

Chinese commentary loves to cast it as credentialed prodigy versus self-taught underdog. Peng is the elite-track engineer: top university, Huawei "Genius Youth," the polished maker-celebrity. Wang Xingxing is the scrappy autodidact who openly admits he was a weak student — he says he passed English only three times in high school and was rejected by Zhejiang University over it. Their strategies differ too: Unitree runs a "profit first, then scale" playbook and has reportedly reached profitability, while AgiBot bets on "scenario, then scale, then profit." In a sign of how institutional they've both become, the two were named vice-directors of China's official humanoid-robot standards committee in late 2025 — the people who will literally help write the rulebook we described in China's plan to give every robot a digital ID.

What he gets right — and the risks

Strip away the viral fame and the case for Peng Zhihui is straightforward, and so are the worries:

• He can actually build it. AgiBot is run by engineers who have personally shipped full-stack hardware and AI, and it shows in the speed from prototype to a 10,000-unit factory.
• Platform, not just product. Open-sourcing the Lingxi X1, the GO-1 model and the AgiBot World dataset is a deliberate play to make AgiBot the layer other robots are built on.
• A magnet for capital and attention. A founder who can trend on Bilibili is a founder who can recruit, raise and sell — an underrated advantage in a crowded field.
• The risks are real. AgiBot lost a co-founder and senior leaders in a round of departures in 2025; it is scaling production faster than profit, in a market many (including us, in our bubble piece) think will see a brutal shakeout; and its listed-company maneuver has drawn regulatory scrutiny. Shipping the most robots is not the same as making robots that reliably hold down a job.

The quick facts on Peng Zhihui

Frequently Asked Questions