AGIBOT's 24-hour convenience store test: what 6 humanoid robots can do
One store, 24 hours, 6 robots, and real customers. AGIBOT ran a fully staffed humanoid convenience store through an entire operating day and livestreamed all of it.
The challenge, called "Robots Never Clock Out," ran November 2 to 3 in China. Six AGIBOT G2 robots handled the full range of convenience store tasks: greeting customers at the door, making coffee, serving hot food, stocking shelves, clearing tables, and collecting rubbish.
The store operated through 4 shifts designed to stress-test different demand patterns: lunch rush, evening rush, overnight low-traffic, and morning breakfast prep.
What they can do
The G2 is AGIBOT's industrial-grade humanoid, built for real environments rather than lab demos. The 24-hour format was designed specifically to show what happens when robots can't clock out and hand off to human staff, requiring multi-robot coordination across the same physical space, across shift transitions, across 24 hours of variable customer density.
The results?
AGIBOT's GO-3 AI model achieved an 85% task success rate and could autonomously identify its own failure causes, modify its approach, and store new knowledge without retraining. And previously, in September, AGIBOT deployed 100 X2 robots across 100 ASD retail stores simultaneously in 3 Chinese cities, running 10 hours a day for 3 consecutive days. That's a commercial deployment at scale.
There's more to see and test
The "Robots Never Clock Out" livestream did not publish autonomy metrics. There's no documented intervention rate, task completion percentage, or failure log for this specific event.
All of that would be interesting data.
But what we did see was impressive ... convenience store operations look deceptively simple: stock shelves, hand over a coffee, greet a customer. In practice they require constant edge-case handling: spills, unusual requests, out-of-stock situations, equipment surprises. The 85% success rate from the GO-3 test in September implies a 1-in-6 failure rate in a controlled setting.
This new 24-hour live store is significantly harder than that.
Why AGIBOT keeps running these challenges
The company is preparing for a Hong Kong IPO, filed in July 2026, and approaching a valuation near RMB 50 billion (roughly $6.9 billion) in a pre-IPO round announced in October.
Public demonstrations at scale, livestreamed, with real customers, are part of the narrative being built for investors and enterprise buyers simultaneously.
AGIBOT's H1 2026 shipments totaled 8,600 units, roughly 35% of global humanoid volume, making it the world's largest humanoid robot manufacturer. The recent "AI Day" added the G2 Max, new embodied reasoning models, and "Worken," a proposed productivity unit for robot labor output. The 24-hour challenge fits the same pattern: each event pushes the public proof of capability forward by one notch, covering convenience stores, cookware chains (100 stores), and theme parks (300-plus robots at Chimelong Spaceship Park).
The challenge format is smart marketing and genuine R&D at the same time. Real customer environments generate real edge cases. Real edge cases generate training data. Training data is what closes the gap between "mostly works" and "ships at scale." AGIBOT already ships at scale.
What they're testing now is whether that scale holds in the messiest environments on the planet.
Sources
Written by
Head of Marketing at NEXT Events & Conferences, responsible for Power Transmission & Distribution Technology Expo and Desalination Technology Expo. And, now, for Humanoid Technology Expo.