Embodied AI's Real Shortage: Organizational Ability to Retain Talent
Over the past six months, funding news has dominated embodied AI: Unitree's STAR Market listing, Zhiyuan's Hong Kong IPO push, Qianxun's 1 billion yuan round, and Star Sea's 2 billion yuan B+ round. Each seems to underscore the industry's rapid rise. But another trend has been overshadowed: a wave of technical and business leaders are leaving the spotlight.
In August, Luo Jianlan, chief scientist at Zhiyuan Robotics, vanished from the company's partner list. Officially a role adjustment, but he removed all Zhiyuan affiliations from his personal page. Luo holds a PhD from UC Berkeley, worked at Google X and DeepMind, and authored SERL, the first robot reinforcement learning system to surpass human level. He joined Zhiyuan in April 2025, built the AgiBot World dataset with over 1 million real-machine operations, and left just 10 days after Zhiyuan announced its Hong Kong IPO.
Also in August, Ding Yan, co-CTO of Luming Robotics, left after less than 280 days. A pioneer in UMI data collection, he had three jobs in 14 months. His departure is attributed to a conflict over technical direction—his UMI body-less data collection approach clashed with other routes. In July, XPeng Robotics core leader Mi Liangchuan and Iron product lead Shi Xiaoxin departed, forcing CEO He Xiaopeng to take over the robotics unit. Earlier, Zheng Cunyuan, head of hardware, had quietly left to start a business. The trio, once called XPeng's Iron Triangle, was completely disbanded.
Similar exits include Zhao Deli, chief scientist at Alibaba DAMO; Zhuang Yuzheng, head of JD's embodied model; Zhang Li, co-founder and COO of Zhuji Dynamics; Xie Junyuan, VP and algorithm head at Qianxun; Wu Changzheng, founder and president of Magic Atom; and Xu Huazhe, co-founder and chief scientist at Star Sea.
A direct cause is the mismatch between capital and technology timelines. Wang Xingxing says embodied AI's ChatGPT moment could be 2-3 years away, or 5-10 years. But capital operates on a shorter clock—over ten companies have started IPO processes this year. When capital demands a three-year exit while technology needs five years to mature, executives are squeezed in the middle.
One investor notes that the industry has been driven by capital and narrative, with valuations doubling monthly and funding growing exponentially. But physics won't accelerate: a robot folding a shirt takes 10 minutes, and it won't become seconds just because the valuation hits 10 billion. The overheated climate has also bred chaos. Shao Tianlan, founder of Mech-Mind, publicly criticized 'assembled' embodied AI companies for using related-party transactions with data centers, local governments, investors, and suppliers to create fake, unsustainable revenue. Luo Jianlan also said benchmark gaming is meaningless. When storytelling pays more than R&D, doers seek alternatives.