SenseTime’s retail technology arm SenseTime Shanhuai has officially launched its SenseMartGo “Shaomai Go” robot convenience store, marking a new phase in the commercialization of embodied AI in offline retail. The first deployments have been rolled out across Shanghai, including Xinzhou Building, Baoshan Xinyefang, and the Baoshan Riverside Scenic Area.
During the recent May Day holiday period, the Baoshan Riverside store demonstrated strong early performance, recording a peak service speed of just 15 seconds per order and an average of 400 orders per day. The figures highlight both the efficiency and growing consumer acceptance of the robotic retail model, which combines automated operations with data-driven store management.
SenseMartGo is built on SenseTime’s five years of accumulated experience in retail AI applications, leveraging advances in computer vision, large-scale retail datasets, and multimodal foundation models. In collaboration with partners including Zero Dimension for customized robotic hardware and Dayao Robotics for data collection systems, the company has developed an embodied AI convenience store solution that goes beyond simple item delivery.
Unlike conventional retail robots that only handle picking and handing over products, SenseMartGo enables end-to-end automation across retail operations. Robots in the system can independently complete order processing, product picking, shelf arrangement, and even cleaning tasks. In addition, the system integrates advanced analytical capabilities that support merchandising, pricing decisions, and restocking strategies, effectively enabling robots to function as autonomous retail operators with both operational and managerial intelligence.
At the core of the system is a “retail brain” built on a three-pillar architecture combining AI capabilities, data assets, and scenario coverage. On the data side, SenseTime has developed a large-scale real-world retail dataset spanning multimodal sources such as video, motion, product, and transaction data. This includes more than 300,000 2D product models, over 100,000 3D product assets, 1.5 million daily transaction records, and more than 8,000 hours of high-density environmental data collection.
On the AI capability layer, the system integrates both classical AI and embodied intelligence. SenseMartGalaxy leverages mature computer vision and multimodal large models, while SenseMartGalaxy-EAI incorporates advanced technologies such as 3D Gaussian reconstruction and world models to enable a closed-loop cycle of perception, understanding, decision-making, and execution for robotic agents.
On the deployment side, the solution spans multiple retail formats, including smart vending units, shelves, and full convenience store environments, allowing flexible adaptation to different commercial spaces and operational needs.
By connecting real-world transactions, environmental data collection, model training, deployment, and continuous iteration, SenseTime has built a self-reinforcing data flywheel. Real retail operations continuously feed high-quality data back into the system, accelerating improvements in both robotic intelligence and operational efficiency. The company plans to accelerate nationwide expansion of SenseMartGo by the end of 2026, aiming to scale embodied AI retail systems across more cities and reshape the future of offline commerce.