A Look Inside the Startup Teaching Robots to Do Blue-Collar Work With Human Teleoperation

Just a 45-minute drive north of central Shenzhen, China, sits IO-AI Tech, a startup charting a surprising new frontier for blue-collar work. Here, human workers strap on the company’s custom VR headsets, handheld controllers, and full-body motion trackers to remotely operate humanoid robots built for roles ranging from factory production floors to neighborhood convenience stores. The company has two core goals: put these robots to work immediately on routine tasks like restocking shelves and sorting parts from storage bins, while collecting rich movement data that will one day let these machines operate fully autonomously.

To demonstrate their technology, the IO-AI team invited me into their offices for a hands-on test. I got to use a custom motion-capturing glove to control 10 distinct robotic hands, each manufactured by a different company. The system translated every tiny twitch of my fingers to all 50 robotic digits instantly, with no noticeable lag.

I’m a little embarrassed to admit that my very first move with this cutting-edge gear was getting all 10 hands to flip the bird. Once I got that juvenile bit of fun out of my system, though, I was blown away by how responsive the system was. Not only did my movements transfer to the robots almost instantly, the feedback worked both ways: I could clearly feel when a small ball was placed into the grip of one of the electronic hands.

The team also let me test a version of the system already being trialed by a major Chinese convenience store chain. Wearing a VR headset and a pair of handheld gripper controllers, I practiced grabbing boxes of medicine off a store shelf and rearranging them. It felt disorienting at first: there was a tiny mismatch between my own movements and the robot’s actions (seen through the headset’s camera feed) that I had to adjust to. But after just a few minutes of practice, I was stacking shelves like a seasoned robot supervisor.

Elsewhere in the facility, I observed full-body teleoperation tests, where operators wore VR headsets and body-tracking sensors that looked straight out of Ready Player One. In one large open room, multiple workers used different IO-AI setups to control small Unitree humanoid robots. One operator walked through a mock apartment set while their robot mirrored every movement step for step right beside them. Seeing the entire space through the robot’s eye-level cameras, the operator worked through the full sequence of motions needed to pull a shirt off a hanger and fold it neatly.

IO-AI builds software that translates a human operator’s natural movements to work across any robot form factor — an increasingly valuable offering right now, as dozens of different humanoid and robotic hand models are already available on the Chinese market. The startup’s algorithms also blend direct human control with limited autonomous adjustment, because humans and robots rarely share the same exact proportions, weight, or center of gravity. Without that small bit of independent correction, robots would regularly lose their balance mid-task.

Shenzhen, home to thousands of hardware and manufacturing firms, is the ideal base for the startup. Co-founder Si Chen explained to me that being located here makes it far faster to build and refine new product prototypes. She also noted that IO-AI is already partnering with a range of local manufacturers eager to automate their most repetitive manual tasks. One partner is Jack Sewing Machines, a Chinese firm that builds apparel production equipment, which is working with IO-AI to train dual-armed robots to handle tasks like ironing shirts. An executive from the company told me these robots can be integrated directly into existing production lines, automating work that is currently done entirely by hand.

Many leading roboticists argue that feeding massive volumes of teleoperation data to AI algorithms will eventually unlock extremely capable, general-purpose robot models. Chen agrees this is the long-term end goal, but she advocates for a more incremental approach to rolling out AI-powered automation.

“It is similar to self-driving cars,” she says, pointing to how autonomous vehicles have gradually rolled out to more environments with steadily increasing levels of independence. “You need this training data that’s more focused on the specific thing you’re trying to address.” She added that robot teleoperation is even gaining traction for training programs at a number of Chinese vocational schools.

China’s manufacturing industry already produces affordable, high-quality robots like Unitree’s humanoids at scale. And if IO-AI Tech is any indication, China’s manufacturing ecosystem may also hold the key to helping AI master work in the physical world.

---

Updated 6/22/2026 10:10 AM EDT: This story has been revised to correct the spelling of co-founder Si Chen.

This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.