Shenzhen Startup Uses VR-Controlled Humanoids for Training Data

Teleoperation as a Path to Robot Autonomy

A startup north of Shenzhen has turned remote robot operation into a job category. At IO-AI Tech, employees don VR headsets, handheld controllers, and motion-tracking equipment to control humanoid robots designed for factory floors and convenience stores, according to WIRED senior writer Will Knight, who visited the facility.

The work serves a dual purpose: the robots perform tasks like shelf stocking and bin picking in real time, while the company captures training data from human operators. That data is intended to eventually enable the robots to work autonomously without human control.

Why It Matters

This approach addresses one of the biggest bottlenecks in robotics: gathering enough real-world training data to make machines reliable in unstructured environments. By having humans teleoperate robots in actual work settings, IO-AI Tech can collect demonstrations of complex manipulation tasks that are difficult to simulate. The model could accelerate the timeline for deploying capable humanoid robots in commercial settings, though it also creates an unusual labor category where workers essentially train their potential replacements.

The Hardware Capital Advantage

The company's location approximately 45 minutes north of downtown Shenzhen positions it in China's manufacturing heartland, where access to hardware components, robotics expertise, and potential deployment sites converge. Shenzhen has become a global center for robotics development, with multiple startups pursuing humanoid platforms and manipulation systems.

The teleoperation setup resembles science fiction depictions of remote work, with operators using their own body movements to control distant machines. This differs from traditional industrial robot programming, which typically requires specialized coding or teaching pendant interfaces.

From Demonstration to Autonomy

The training data collection strategy assumes that observing enough human-controlled operations will allow machine learning systems to generalize and perform tasks independently. This imitation learning approach has shown promise in research settings but faces challenges in achieving the reliability required for commercial deployment.

IO-AI Tech's focus on practical applications like inventory management and material handling targets tasks that are repetitive enough to learn but complex enough that traditional automation has struggled to address them cost-effectively.

Details of the operation were first reported by Will Knight for WIRED, based on his visit to the IO-AI Tech facility.