AgiBot Put Six Humanoid Robots in a Convenience Store for 24 Hours. Two Failed Their Trial

Six AgiBot G2 robots stocked shelves, made coffee and greeted shoppers in a 24-hour Shanghai store livestream. Four passed; two did not.

Editorial illustration of wheeled humanoid robots stocking shelves and making coffee in a bright convenience store at night.
AI-generated Rovivox illustration, not a photo of AgiBot’s G2.

Chinese robot maker AgiBot has put six of its G2 robots to work in a real convenience store for 24 hours, and broadcast the whole thing live.

The challenge, called “Robots Never Clock Out,” ran from 10:00 on November 2 to 10:00 on November 3 China time, according to Interesting Engineering and Humanoids Daily. TechSpot places the store in Shanghai’s Xuhui district.

Four shifts, six jobs
AgiBot split the day into four periods: the lunch rush, the evening rush, late-night duty and breakfast preparation. The robots’ jobs included stocking and organising shelves, making coffee and hot food, greeting customers and clearing rubbish, all while real shoppers moved around them.

The results
By AgiBot’s count, reported by TechSpot and Poland’s Chip.pl, the six robots completed 1,730 tasks, including 683 customer greetings, 369 item retrievals and deliveries, and 147 coffees.

But only four of the six passed the company’s assessment. The shelf-patrol robot reportedly could not keep up with the store’s turnover, and the cleaning robot, on night duty, met a full bin, left its route and wandered out of the store, according to TechSpot, which cites ITHome.

Chip.pl adds useful context: almost 40% of the logged tasks were greetings and customer service, which are much easier than physically grabbing products.

What the livestream doesn’t show
Several outlets make the same point. The 24 hours describe the event, not the robots. AgiBot did not say whether individual machines needed charging, maintenance or human help, and it published no intervention policy or scoring method. There is also no data on whether the robots matched human staff on speed or cost.

Why it matters
Shops are a harder test than a factory station. Layouts change, customers get in the way, and requests are unpredictable. A day-long public stream, failures included, is more honest than a polished 30-second clip. The next step that would really count is the boring one: intervention logs and cost per task.