Lian Wenzhao's (连文昭) résumé reads like a wish list for any roboticist. After a PhD at Duke University, he joined Vicarious, a Silicon Valley artificial-general-intelligence company; then Google X's robotics project Intrinsic; and then Figure, one of the world's best-known humanoid-robot startups. All three are among the most elite institutions in embodied AI. In 2023 he returned to China to found Yuanluo Tech (源络科技) and also became a professor at Shanghai Jiao Tong University's School of Artificial Intelligence.
Yet in an industry where everyone is screaming "faster," his best-known position is the opposite: slow down.
For readers outside China: "embodied AI" means artificial intelligence given a body — robots that perceive and act in the physical world. The field has become one of China's hottest investment and policy themes in 2025–2026, which makes a calm voice arguing for restraint worth hearing.
Key takeaways
- Lian spent years at Vicarious, Google X (Intrinsic), and Figure before founding Yuanluo Tech (源络科技) in 2023.
- He frames robotics deployment through an "impossible triangle" of reliability, generalization, and speed — and chooses to sacrifice speed first.
- Yuanluo trains on "broad lab" scenarios: long-sequence, high-flexibility, high-precision science experiments.
- China's National Development and Reform Commission (发改委) recently warned against blind跟风 (following the crowd) and rushed pile-ons.
- His rule for any founder: "Can it deliver?" beats any demo that fools an audience but not a daily user.
The cold keyword that became the hot track
Two years ago, when Lian mentioned "intuitive physics" on a podcast, it was a niche phrase. Today, reborn under the name "world model," it is one of the hottest startup directions in the field. Back then he used the "impossible triangle" to describe why robotics is hard: among reliability, generalization, and speed, he chose not to over-demand speed and to protect the first two.
That choice looked costly at the time. Others were shipping demos — folding-clothes and coffee-pouring clips flooding feeds — while he ground away at unseen foundational capability. Two years on, the wind has shifted to validate restraint: the market now asks "can it hold a full shift reliably?" rather than "can it backflip?"
The hardest road: the laboratory
Yuanluo does not demo folding clothes or pouring coffee. It plunges into "broad lab scenarios," using long-sequence, high-flexibility, high-precision scientific-experiment tasks as the robot's training ground. Why there? A lab offers three advantages. The tasks are hard enough to force out the robot's true capability ceiling. The standards are rigid enough that right and wrong are unambiguous — no fudging. And the value is high enough: scientific experiments are repetitive and tedious, human labor is expensive, and the substitution need is real.
His bet is heating up. As AI changes the paradigm of scientific discovery, the bottleneck has moved to the "wet lab" — the hands-on steps that must touch real reagents and materials. The robot, he argues, may be the physical interface that closes the loop between dry (computational) and wet (experimental) science.
A cool head's unwelcome advice
Last month the industry was euphoric: the World Robot Conference and a humanoid-robot games event ran back to back, and Unitree (宇树科技) listed on the Star Market (科创板). Then the National Development and Reform Commission (发改委, the NDRC — China's top economic planner) urged preventing blind imitation and rushed pile-ons. Lian's view lines up closely with that warning.
He does not deny progress — this year's competing robots began using vision to judge obstacles and decide how to step, without an operator issuing step-by-step commands. But he cares more about what is unsolved: is reliability enough, can it run fault-free long term, and how fast can it recover when something breaks. What is rarest in the noise is the courage to admit "not ready yet."
What an ordinary person can take from it
The reporting extracts three lessons that travel well beyond robotics. First, in a hype cycle, choose the hard problem; easy problems are crowded and their payoff window is shortest, while hard problems have fewer competitors and compound. Second, guard your own "impossible triangle" — with limited resources you must rank priorities, and reaching for everything usually means achieving nothing. Third, test yourself by "can it be delivered": a demo can fool spectators, but not the person who has to use it every day.
The line worth keeping: when the industry is hottest, the most valuable company is not the fastest runner but the one that knows exactly where it still falls short.
Honest limitations
This article is a profile built on a Chinese tech-outlet interview/narrative and on public facts about Lian Wenzhao's (连文昭) career; the biographical sequence (Duke, Vicarious, Google X/Intrinsic, Figure, founding Yuanluo Tech in 2023, professorship at Shanghai Jiao Tong University) is reported but not independently re-verified here against primary records. Quotes and the "impossible triangle" framing are attributed to Lian via the outlet and should be checked against his own published talks before citation. The NDRC warning is reported secondhand; the precise wording of the advisory should be confirmed from the original government statement. The claims about robot capabilities at recent conferences are the outlet's observations, not benchmarked results. No investment or career advice is intended.
Sources
Based on reporting by the Chinese tech outlet Naojiwang (脑机派对), 2026-09-03, drawing on an interview/narrative about Lian Wenzhao (连文昭) and Yuanluo Tech (源络科技), and on statements attributed to China's National Development and Reform Commission (发改委).