Six AgiBot G2 humanoid robots ran an operating convenience store in Shanghai's Xuhui district for a full 24 hours, completing 1,730 tasks and drawing more than 2 million livestream viewers, according to The Robot Report. Only four of the six machines passed AgiBot's final performance assessment.
The trial, called "Robots Never Clock Out," ran from 10 a.m. on November 2 to 10 a.m. on November 3. Each wheeled humanoid was assigned a distinct role: customer assistance, shelf inspection and restocking, coffee preparation, checkout and hot-food service, or store cleaning. AgiBot's tally breaks down to 683 customer-service interactions, 369 product pickups and deliveries, 147 coffee orders, 126 breakfast servings, 277 shelf-restocking tasks and 128 table-cleaning assignments.
The 24-hour window exposed the robots to four distinct operating conditions: the lunchtime rush, the evening peak, overnight quiet hours and the following morning's breakfast surge, rather than a single rehearsed demo. AgiBot said the exercise combined its robotics hardware, its Genie Master embodied AI agent framework and the Embodied Execution Protocol, which handles task coordination and real-time visual adaptation, in one continuous working environment.
Why a retail floor beats a lab demo
Humanoid makers have increasingly swapped staged showcases for live commercial settings to prove reliability, since buyers and investors now discount choreographed demos. AgiBot's choice of a working convenience store, rather than a warehouse or factory floor, also targets a market segment few humanoid developers have tested: direct customer interaction under unscripted conditions, including handling impatient shoppers and unexpected requests.
The two robots that failed the final assessment are a reminder that extended, varied-task autonomy is still the industry's hardest problem. AgiBot said its evaluation criteria covered duty time, task completion rate, customer feedback, response times and handling of unexpected situations, a broader bar than the single-skill benchmarks most humanoid demos use.
What happens next
AgiBot said it will study which capabilities and operating practices from the trial can be transferred between tasks and replicated across different retail and service settings, feeding that data back into further real-world testing. The company has positioned retail and service work as a parallel track to its industrial and logistics humanoid lines.
The trial adds to a wave of Chinese humanoid field tests this year aimed at proving commercial readiness ahead of the Lunar New Year shopping season, when retailers typically evaluate staffing technology for the following year's rollout.