Robot Shopkeepers in Hefei: A Real-World Test for Humanoid Retail
On August 26, 2026, in the southern section of Lvjie Street in Hefei, a wheeled humanoid robot named 'Xiaomai' was busy taking orders, dressed in a white T-shirt and an orange tennis cap. After customers scanned a QR code or placed orders by voice, the robot swiftly turned to shelves, precisely identified and grasped products, and delivered a bottled drink to the pickup port in about 20 seconds. This was one of the first four 'robot shops' launched by Hefei Guoxian Holdings in partnership with Zero Robotics, marking the first large-scale attempt at an embodied-intelligence unmanned retail project in Anhui Province.
Zero Robotics founder Min Yuheng emphasized that the key to moving robots from labs to real commercial scenarios lies in their ability to operate reliably amid continuously changing environments. Product dimensions, materials, shapes, and placement vary constantly, demanding high generalization, real-time error correction, and failure recovery capabilities. The four shops were intentionally placed in bustling areas like Lvjie Street and Hefei 1972, rather than standardized spaces, to expose robots to real customer flow and unexpected situations. According to Zero Robotics staff, after months of testing, the overall order fulfillment rate reached 99%, and during earlier trials in Shanghai, a single store achieved weekly revenue exceeding 21,000 yuan.
Behind these robot shopkeepers lies Hefei's systematic industrial strategy. In Yaohai District, a 1,600-square-meter public service platform for intelligent robots employs about 40 data collectors (born after 2000) to conduct 'training' for various robots. The platform replicates 33 real scenarios covering industrial manufacturing, commercial services, and household life, allowing robots to repeatedly practice skills like grasping, moving, and recognition. Xu Bin, deputy general manager of platform operator Hefei Guoxian Holdings, explained that the biggest challenge for commercial humanoid robots is the scarcity of scenarios and data. Previously, companies building their own training environments faced high costs, long cycles, and slow results. Now, robots can undergo realistic training on the platform before direct deployment to street shops, with physical interaction data feeding back into model iteration—forming a closed loop of 'scenario-data-model-ability'.
In July of this year, the 'Universal Grasping Dataset' collected by Hefei Embodied Data Operations Management Company completed property rights registration at the Anhui Data Exchange. This dataset covers full-process operations such as object grabbing, placing, and arranging, becoming the province's first multimodal grasping dataset for embodied intelligence. Hefei's robot industry layout has been long in the making. In March 2025, the city launched the '6128' project, planning a 10-billion-yuan future-industry fund, investing 2 billion yuan over three years for public platform construction, and opening supply-chain catalogs of 60 vehicle enterprises within a year, integrating with the multimodal 'smart brains' of Huawei and iFlytek. To date, Hefei has attracted nearly 200 companies in the intelligent robot industry chain.