AgiBot's 6 Genie G2 Robots Work 24-Hour Live-Streamed Shifts at a Convenience Store

AgiBot has started livestreaming again, and this time the setting is a convenience store. At 10 a.m. today, six Genie G2 robots entered an operating convenience store in Xuhui District, taking on tasks such as greeting and reception, coffee preparation, hot food service, picking and restocking goods, and table cleaning, and they will be under real-time online "supervision" from the entire internet for 24 hours. The 24 hours at the convenience store cover four shifts—the early-morning breakfast rush, the midday peak trial shift, the evening peak stress test, and late-night independent duty—which is no small challenge for the robots. AgiBot hopes that through a livestream in a real operating scenario, it can verify the robots' continuous operation and autonomous service capabilities and push humanoid robots from demonstrations toward real on-the-job work. Six "store clerks" work nonstop as tricky tests come one after another. With a small footprint and standardized processes, the convenience store is one of the most common commercial scenarios in the city. But for robots, being competent in the work there is not easy. In the livestream footage, the first customer to enter the store had just reached the door when the reception robot "Ahao" proactively stepped forward. "The convenience store is like your home—pick any product you like, and you can chat even if you don't buy anything." While greeting the customer, Ahao chatted with them and used a smooth set of "compliment combos" to liven up the atmosphere. "One vote from you, one vote from me, and Ahao gets confirmed after probation and debuts," making the customer laugh continuously. The six robots each perform their own duties. After receiving an order for a bottle of milk, the picker "Huobao" went to the shelf and precisely grasped the milk with its gripper; the restocker "Juanwang" patrolled the store and proactively restocked wherever it saw an empty shelf. The customer then went to the checkout counter and ordered a cup of coffee and a steamed bun from the tablet. "Kashao," wearing a bow tie, said "May coffee start your good mood" while skillfully beginning to prepare it; the hot food station worker "Dingjiang" opened the glass cabinet door holding breakfast items, took a meat bun from the steamer, and put it into a kraft paper bag. The cleaning worker "Lengjie" patrolled the tables in the store and, while grumbling, cleaned the tabletops. As of 2:30 p.m. yesterday, the six probationary "workers" had been on duty for 4.5 hours and completed a total of 337 tasks, including receiving customers 116 times, picking up and delivering goods 84 times, making coffee 30 times, providing meal service 40 times, restocking and patrolling 27 times, and cleaning tabletops 40 times. When they slip up, they enter the "wrong-answer collection," and the robot collective continues to evolve. Being able to stay on duty continuously in a convenience store depends not on a single-point demonstration but on models, data, and the post-training system. Just a few days before the 24-hour convenience store challenge livestream, AgiBot released core technologies such as the GO-3 foundation model and the "Genie Master" embodied intelligent agent, showing that the robots can already complete long-horizon tasks such as making jianbing guozi and inserting and removing LEGO bricks. AgiBot has built a distributed mixed-environment post-training system, relying on hundreds of real machines and advancing through a combination of physical simulation and world models. However, even after tens of millions of training runs offstage, it is still impossible to exhaust all the difficulties encountered in real scenarios. How can they be made unafraid of new environments? This is precisely the starting point for AgiBot's previously launched embodied intelligent agent framework "Genie Master." As the shared intelligent agent "brain" of the six robots, it is responsible both for scheduling tasks and for connecting execution results with subsequent learning. "Genie Master" summarizes successful and failed work trajectories every day and stores the relevant experience in long-term memory, directly calling it when encountering similar tasks; the updated policy then returns to the environment for verification, forming an "exploration—feedback—improvement" loop. The two paths of "remembering experience" and "mastering skills" work together, so that each round of work provides feedback for the next round of learning, and the robots thereby "learn new skills while on duty." An even more impressive technical foundation is the new paradigm for collective experience learning proposed by AgiBot together with Tsinghua University and Infinigence AI, which has completed large-scale validation in a cluster of 96 robots. Simply put, the successful experience that one robot figures out in practice can be accumulated and shared and reused by the entire robot collective. Under this paradigm, robots show a trend of "the more, the stronger": as the number of robots participating in training increased from 16 to 96, the success rate rose from 86.7% to 98.3%. Robots "taking up real jobs" has already come to our side. AgiBot is a company that likes to speak through livestreams. Last July, its self-developed A2-W robot launched the world's first "factory work livestream" at Fuling Technology's Mianyang factory, with zero errors throughout the 3-hour process; last August, Expedition A2 launched a 24-hour "Summer Citywalk" downstairs from the company; in November, Expedition A2 went on an "expedition" of 106 kilometers, walking from Suzhou's Jinji Lake to Shanghai's North Bund, going through 15 battery swaps over a full 56 hours. This June, AgiBot launched the internet's most hardcore real factory livestream at the Nanchang Longcheer Technology factory, with all eight Genie G2 robots on duty, strictly following the factory standard of 8 a.m. to 7 p.m. and working continuously on real production lines for six full days, completing quality inspection processes such as tablet interfaces and audio radiation, with a work success rate as high as 99.99%. From walking alone to interacting with multiple people, from working at fixed workstations to working in rich scenarios, robots are increasingly able to take on people's work. Although this 24-hour convenience store challenge livestream was not fully open to the public and customer flow was arranged in single digits, the unstoppable trend of robots taking up jobs can still be seen. Shopping malls, amusement parks, gas stations... if you take a walk on the street now, you may find that robots have already integrated into our lives.