Six AgiBot Genie G2 Humanoid Robots Work 24-Hour Shifts at a Shanghai Convenience Store

At 10:00 on November 2, in an open convenience store in Xuhui, Shanghai, six special "new employees" clocked in for work. They are six AgiBot Genie G2 humanoid robots, serving respectively as greeters, baristas, stockers, and cleaners, working a full day continuously in this 24-hour convenience store—taking orders, making products, restocking, and cleaning—taking on the "Robots Can't Stop Working" challenge and being "supervised" by viewers across the internet via livestream. The "new employees" worked four shifts back-to-back as tricky test questions came one after another. On the badges of the six robot "new employees" were written their respective positions: Ahao stood at the entrance, responsible for greeting and customer inquiries; Huobao and Juanwang were a stocking pair, one responsible for retrieving and restocking goods, the other for patrolling the shelves; Kashao was the dedicated "barista" at the coffee station; Dingjiang guarded the checkout counter and was also responsible for prepared food service; Lengjie took on the store's table cleaning work. The six positions covered multiple links in the convenience store, from receiving customers to merchandise management, food and beverage service, and daily cleaning. Beyond the positions, there were also scheduling and assessments. The six robots needed to go through different work periods including the lunch peak, evening customer flow, late-night duty, and the next morning's early peak. Their on-duty hours, business completion volume, customer evaluations, as well as task success rate, response speed, and exception-handling capability would all be included in the assessment, becoming the basis for whether they could "become regular employees." On the day of the challenge, two specially invited experience officers arrived with test questions and conducted two rounds of store inspections, setting various "tricky problems" for the robots from the customer's perspective: nearly identical sugary and sugar-free colas—with a single phrase "get the sugar-free one," would it pick the wrong one; completely messing up the gum shelf—could it put each box back in place; standing in the middle of the aisle as a "human roadblock" and then placing a box to block the way—would it choose to go around or directly bump into it. Facing these tests, the robots also had their own ways of responding. Before picking up a cola, they first understood the customer's needs through voice recognition, then confirmed the product visually; when tidying a messed-up shelf, they did not rely on preset labels but used visual inventory to put items back grid by grid; when encountering a pedestrian, they stopped first and then found a path around, and when a box blocked the way ahead, they would also re-plan their route. A cup of coffee, a restock—small actions yet highly difficult. With a small area and relatively standardized processes, convenience stores are common commercial scenes in cities. But for robots, truly being competent in such work is not easy. For example, hanging a bag of Skittles on a hook—if the angle is slightly off, it may clip the neighboring product; picking up a piece of golden cake—if the force is slightly improper, the result may be dropping or crushing it. In the laboratory, lighting, item positions, and operating procedures can all be set in advance; but in a real store, customers walk through at any time, and the positions, orientations, and packaging states of products are constantly changing. Even when picking up the same item, the conditions faced each time may differ. One successful grasp does not mean the next one can be completed the same way. Facing these tasks, the robots need to continuously adjust their operating methods according to the on-site situation. For products customers casually put in the wrong place, the robots use real-time visual inventory to find their correct positions and then put them back grid by grid; when discovering a gap on the shelf, the restocking robot can automatically trigger the restocking process without manual reminders; facing stains on the counter, it also needs to distinguish between the product's own pattern and traces left by spilled liquid, and then call cleaning tools for handling. Failures are recorded in the "mistake collection," allowing the robots to continuously accumulate experience. During real business operations, the robots did not always complete tasks smoothly and occasionally failed during work. But more noteworthy than whether a single task succeeds is: can the robots identify problems, correct errors, and turn the handling process into experience for the next operation? This is also the starting point for AgiBot's previously launched embodied Agent framework Genie Master. As the shared Agent "brain" of the six robots, it is responsible for task scheduling and also connects the robots' execution results with subsequent learning. On-the-spot correction is not enough; the robots also need to turn newly acquired experience into reusable capabilities. Genie Master summarizes successful and failed work trajectories every day, stores the relevant experience in long-term memory, and can call it when encountering similar tasks; effective adjustment processes will form training data, feeding back into the continuous optimization of motion strategies. The two paths of "remembering experience" and "training skills" work together, allowing each round of work to provide feedback for the next round of learning. In this way, the robots can accumulate experience in actual work and gradually improve their task completion capabilities. This mechanism has already been applied in store operations. Over the past period, the Genie G2 has continuously improved on multiple complex and delicate tasks; when the store is idle late at night, the robots will also actively practice skills. Right now, this convenience store challenge is still ongoing. After two shifts of late-night duty and the next morning's early peak, the six Genie G2 robots will face their final assessment. Can they successfully "become regular employees"? This continuous 24-hour convenience store practical battle will also provide a stage-by-stage answer sheet on the real operational capabilities of embodied intelligence.