In Shenzhen, China's hardware manufacturing hub, a new form of blue-collar work is emerging: teleoperating humanoid robots to gather training data for AI-powered automation. IO-AI Tech, a startup located about 45 minutes north of downtown Shenzhen, has developed systems that allow workers to control multiple humanoids simultaneously using VR headsets, handheld controllers, and motion-tracking gear. The company's dual goal is to have robots perform useful tasks immediately—such as stocking shelves in convenience stores or picking items from bins—while amassing the data needed for eventual autonomous operation.
Teleoperation as a Workforce Solution
At IO-AI Tech's facility, workers wearing virtual reality headsets and body-tracking sensors—reminiscent of Ready Player One—manipulate humanoid robots from Unitree. One demonstration showed an operator marching alongside a Unitree robot that mirrored the person's movements within a mock apartment, performing actions like removing a shirt from a hanger and folding it. The operator viewed the scene through the robot's eye-level cameras.
The author of the original report, Will Knight, tested a system being piloted by a Chinese convenience store chain. Using a VR headset and a pair of grippers, he attempted to pick up boxes of medication from a shelf. After adjusting to a slight delay between his movements and the robot's, he was able to stack shelves effectively. The company also let him wear a custom motion-tracking glove that instantly transferred finger movements to 10 humanoid robotic hands from different manufacturers, all of which flipped the bird in unison.
Training Data for Autonomous Future
Si Chin, cofounder of IO-AI Tech, explained that the company's algorithms combine human control with some autonomy to compensate for differences in body shape, size, and weight between operator and robot. Without this ability, the robot might lose balance. The training data gathered from teleoperation is intended to eventually unlock general AI models capable of independent action. However, Chin advocates an incremental deployment approach, comparing it to the development of self-driving cars: "It is similar to self-driving cars. You need this training data that’s more focused on the specific thing you’re trying to address."
Collaboration with Manufacturers
IO-AI Tech's location in Shenzhen—home to thousands of manufacturers—enables rapid prototyping and refinement. The company is already working with local industrial firms. One notable partner is Jack Sewing Machines, a manufacturer of clothes production equipment. Together, they are training two-armed robots to perform tasks like ironing shirts. These robots could be integrated into existing production lines to automate work currently done by hand, according to an executive from Jack Sewing Machines.
| Application | Task | Partner/Client | Current Stage |
|---|---|---|---|
| Convenience store | Picking medication boxes from shelves | Unnamed Chinese chain | Pilot testing |
| Apparel manufacturing | Ironing shirts | Jack Sewing Machines | Training |
| Simulated home | Removing and folding a shirt (Unitree robot) | IO-AI Tech demo | Internal demo |
Incremental Path to AI Automation
While some roboticists believe that massive teleoperation datasets will quickly produce general-purpose AI, Chin takes a pragmatic view. She noted that robot teleoperation is already gaining traction in Chinese vocational schools, suggesting a skilled workforce pipeline. China's manufacturing strength, as seen in the affordable, high-quality robots from Unitree, combined with companies like IO-AI Tech, may accelerate the mastery of AI in the physical world.
For enterprise technology leaders, the Shenzhen model highlights a viable near-term strategy: using human-in-the-loop teleoperation to both solve current labor shortages and create the training foundation for future autonomous systems. The partnerships with existing manufacturers, such as Jack Sewing Machines, underscore that this approach is not speculative but already embedding into production environments. Decision-makers evaluating automation investments should consider whether teleoperation-as-data-collection can shorten the ROI timeline for their specific use cases.
Sources:
