AGIBOT Robots Complete 1,730 Tasks in 24-Hour Retail Trial
Six AGIBOT G2 humanoid robots performed over 1,700 customer service and operational tasks during a continuous trial in a Shanghai convenience store, demonstrating embodied AI's ability to handle real-world retail environments.
Six AGIBOT G2 robots completed 1,730 tasks in a 24-hour trial at a Shanghai convenience store. The wheeled humanoid robots operated from 10 a.m. on November 2 to 10 a.m. on November 3 in an experiment organized by the Chinese robotics company AGIBOT. Their responsibilities spanned customer assistance, shelf inspections, restocking, coffee preparation, checkout, hot food service, and store cleaning.
The trial covered four distinct operating periods: the lunchtime rush, evening peak hours, overnight operations, and the following morning’s breakfast rush. This schedule tested the system's endurance across varying customer traffic levels. Four robots passed the final assessment, while two were placed in an observation group for further improvement.
Operational Performance Metrics
The robots' 1,730 completed tasks broke down into specific functions. Their performance data provides a concrete measure of their operational capacity in a live setting.
| Task Category | Number Completed |
|---|---|
| Customer-service interactions | 683 |
| Product pickups and deliveries | 369 |
| Coffee orders | 147 |
| Breakfast servings | 126 |
| Shelf-inspection and restocking tasks | 277 |
| Table-cleaning assignments | 128 |
Two robots failed during the trial. One cleaning robot deviated from its path after encountering a full trash bin and failed to return to its position. A separate shelf-inspection and restocking robot struggled to consistently identify and replenish products as merchandise was sold.
Technical Capabilities Demonstrated
The trial was designed to test the robots' ability to handle unexpected situations and adapt without explicit human prompting. They used real-time visual information to identify and return misplaced merchandise to correct shelf positions. During restocking, they detected shelf gaps and replenished products autonomously. In cleaning tasks, they distinguished spills from countertop patterns and selected appropriate tools.
AGIBOT subjected the robots to specific, challenging tests. These included distinguishing between regular and sugar-free Coca-Cola cans, rearranging chewing gum packets, and navigating aisles blocked by a person and a box. The systems demonstrated an ability to adapt to changing conditions like varying product orientations.
AI Coordination Framework
AGIBOT uses its Genie Master embodied AI agent framework and an Embodied Execution Protocol to coordinate tasks. This system linked task planning, execution, and feedback, organizing work based on individual robot capabilities and live store conditions. The Genie Master framework enabled error correction and supported learning from operational experience.
The company said the trial showed how embodied AI could move beyond controlled demonstrations and operate in the changing conditions of a live, customer-facing environment. The AI retained experience from both successful and unsuccessful task executions in long-term memory. Effective adjustments were converted into training data to refine the robots' action policies for future operations.
AGIBOT will continue testing robots across additional roles and real-world environments. The company plans to use operational feedback from trials like this one to improve robotic reliability, task execution, and overall adaptability.
