Chen Jianyu was still a PhD student at Berkeley when he first started asking why robots could not simply learn to do useful things. Years later, as a professor at Tsinghua, he decided the answer was not another paper but a company.
The company is Robot Era, and its pitch is unfashionably literal: build the whole machine — the body and the mind — and then send it to an actual job.
Inside the firm's Beijing workshops, human-shaped robots are trained not to dance but to pick, place, and sort. That choice, more than any single spec, tells you what Chen is after.
A professor who refused to pick a side
Robot Era (星动纪元) was founded in August 2023 and is, by the company's own description, the only embodied intelligence (具身智能) venture in which Tsinghua University holds equity. Chen, born in the early 1990s, was admitted to Tsinghua's precision instrumentation department in 2011 and later earned a PhD at UC Berkeley, where he studied under Masayoshi Tomizuka, a pioneer of model predictive control.
He returned to China around 2020 and joined Tsinghua's Institute for Interdisciplinary Information Sciences as an assistant professor — reportedly one of the university's youngest PhD supervisors, recruited at the invitation of Turing Award winner Yao Qizhi (姚期智). His research circled the same question from every angle: bipedal and quadruped robots, self-driving cars, robot arms, and grippers. The throughline was always the same — build general-purpose robots with general-purpose intelligence.
Building both the body and the brain
Most robotics startups pick one end of the problem. Chen's team picked both, a strategy the company calls "software-hardware integration" (软硬一体).
- A native robotics model. In late 2024 the company released ERA-42, an end-to-end native robotics large model (大模型), paired with a five-finger dexterous hand, XHAND1, said to handle more than a hundred delicate tasks.
- Full-size hardware. Robot Era has shown a full-size humanoid, the STAR L7, and a service robot, the Q5, rather than limiting itself to research arms.
- One delivery, not two. As Chen put it in a China News Weekly interview, a robot without a brain is scrap metal, while a brain without a body is hardly a robot. The point is that the value only shows up when the two are shipped together.
That vertical stance is expensive and slow at the start. Chen's counter is that the commercial payoff in embodied intelligence (具身智能) only arrives as a coordinated whole, not as a clever model bolted onto someone else's chassis.
From demo to the sorting line
The practical test is a warehouse, not a stage. Robot Era has leaned into logistics, and in 2026 its humanoid, the STAR M7, began working parcel-sorting shifts alongside SF Express (顺丰). Co-founder Xi Yue (席悦), a middle-school classmate of Chen's whom he recruited from Singapore, has said the company's main strategy this year is logistics — the fastest route from prototype to scaled orders.
The funding has followed the plan. Robot Era closed a near-500-million-yuan A round (about US$70 million, HK$550 million) in July 2025. By April 2026 it had completed a further round of more than US$200 million, and Chinese business press tallied roughly 3.5 billion yuan (about US$493 million, HK$3.85 billion) raised across three rounds between late 2025 and early 2026. Backers range from Sequoia China and IDG to industrial partners including SF Group, Geely Capital, BAIC, and Haier. The company also claims that nine of the world's ten largest tech firms by market value are customers — a statement worth reading as a sales claim, not an audited fact.
Why a professor started a factory
Chen has described his shift from lab to company as the only way to close the loop. The university gave him room for zero-to-one breakthroughs; turning that into something ten-times larger, he argues, requires the feedback of real deployments. Robot Era's early investors reportedly queued up before the company formally existed — a sign of how hot the humanoid category had become — yet the team resisted the easy path of building a robot that performs for cameras.
The bet is that dull, repeatable tasks — moving boxes, sorting parcels — are where humanoids first earn their keep, even if they are less photogenic than a backflip.
Trend reading
According to Li An, chief scientist at BrainNet (脑机网), the divide between a robot that demos well in a video and one that survives a real shift is still where most embodied intelligence (具身智能) teams get stuck, and the teams that win will be the ones measured by uptime rather than footage.
Honest limitations
The core facts here — founding date, Chen's Tsinghua role, the ERA-42 model and XHAND1 hand, and the SF Express logistics push — come from Robot Era's own site and from reporting in China News Weekly and China Economic Net. Funding totals are drawn from Chinese business press and should be treated as reported figures, not audited results; the "nine of ten" customer claim is the company's own statement. A crowd-sourced Baidu entry was used only to cross-check the product timeline, not for any factual claim. What remains unknown is true unit economics, how many robots are actually in paid daily operation, and whether the logistics deployments will survive contact with full-scale commercial contracts.
What readers can do now
- If you track robotics commercially, watch Robot Era's SF Express deployment for throughput and failure-rate numbers, not launch photos.
- If you build on open models, follow ERA-42's open-source releases to see how a Chinese lab trains whole-body control.
- If you are a founder, read the China News Weekly profile for Chen's own words on why he kept hardware and software under one roof.