Humanoid Robots Fail Convenience Store Trial After Stumbling on Job Tasks

Six humanoid robots finished a 24-hour work trial at a convenience store, but two failed their probation because they could not keep up with the work. The test highlighted the current limits of automation when machines are placed in real-world retail jobs.

Robot manufacturer Agibot placed its G2 humanoid machines in an operating convenience store in Shanghai's Xuhui district from November 2 to 3. The broadcast, called "Robots Never Clock Out," tested the machines across four periods that included the lunchtime rush, evening crowds, overnight duty, and breakfast preparation. Their duties involved stocking shelves, preparing coffee and hot food, greeting customers, and clearing rubbish. Testers also added difficult tests, such as asking the robots to tell sugary Coke from sugar-free Coke, placing obstacles in their paths, and shaking containers while a machine tried to pick up food.

By the end of the shift, the robots completed 1,730 tasks, which included 683 customer greetings, 369 item retrievals and deliveries, and 147 coffees. However, only four machines passed the assessment. The cleaning robot and another machine assigned to patrol the shelves fell short. According to ITHome, the shelf-patrol robot lacked initiative in inspecting and restocking goods and failed to match the store's turnover pace, meaning its on-duty enthusiasm needs improvement. Meanwhile, the cleaning robot encountered a full trash can on night duty, which caused it to leave its designated route, run outside the store, and stray from its post.

The report did not specify whether individual machines needed charging breaks, maintenance, or human assistance during the 24 hours of continuous operation. It also remains unknown whether the robots matched human workers for speed and cost. A separate report from January noted that humanoid robots generally work about half as fast as human counterparts.

Summary

The convenience store trial shows that while humanoid robots can handle basic retail tasks, they still struggle with the unpredictability and pace of real-world jobs. These limitations highlight the gap between automated demonstrations and reliable everyday labor.