en.Wedoany.com Reported - SenseTime's subsidiary SenseMart officially launched its SenseMartGo robot retail solution in May 2026, with the first three directly-operated stores landing in Shanghai's Xinzhou Tower, Baoshan Xinyefang, and Baoshan Riverside Scenic Area, and plans to complete a 100-store network by the end of 2026. This launch marks SenseTime's advancement of artificial intelligence technology from laboratory demonstration to large-scale commercial application, as embodied intelligent convenience stores officially enter the replicable and scalable physical retail track.
During the May Day holiday period, the Baoshan Riverside Scenic Area store delivered operational results of "fastest 15 seconds per order, average 400 orders per day." The three stores have achieved normalized operations, with a full-cycle average delivery time of 29 seconds, 24-hour operational stability reaching 99%, and a fully standardized solution supporting rapid store setup within 7 days, adaptable to scenarios such as office buildings, industrial parks, scenic spots, and communities. This solution breaks the limitation of traditional smart retail cabinets that can only sell standardized goods, fully covering standard products like beverages and snacks, while also supporting the preparation of non-standard items such as coffee, ice cream, hot breakfast foods, and popcorn, capable of carrying over 300 SKUs within a minimal space of 15 square meters.
SenseMartGo robot retail is a complete embodied intelligent convenience store solution, built upon SenseTime's five years of technological accumulation in the retail industry, relying on computer vision technology, massive retail data assets, and multimodal large model capabilities, jointly created with a robot cooperation ecosystem including customized hardware from Zero Dynamics and the robotics data acquisition system from Daxiao Robotics. Unlike single-function vending robots on the market that can only pick and deliver items, this solution achieves full coverage of retail task types. The robot not only autonomously completes basic tasks such as order taking, picking, restocking, and cleaning, but also possesses data analysis and operational decision-making capabilities for product selection, pricing, and replenishment, positioned by the internal team as an embodied intelligent robot system that "can sell, understands operations, and makes money."
At the technical foundation level, SenseTime has built a three-in-one full-stack system of "AI Capability × Data Assets × Scenario Coverage," creating a core platform named "Retail Brain." On the data asset side, it has accumulated over 300,000 2D product models, over 100,000 3D product asset libraries, as well as over 1.5 million daily transaction order data and 8,000 hours of high-density environmental data collection. On the AI capability side, it has constructed a "Classic AI + Embodied" dual-engine architecture. SenseMartGalaxy is based on mature computer vision and multimodal large model capabilities, while SenseMartGalaxy-EAI further integrates cutting-edge technologies such as embodied brain 3D Gaussian reconstruction and world models, endowing the robot with a closed-loop capability of "perception—understanding—decision-making—execution." Since 2021, SenseTime has successively completed static image perception, dynamic video analysis, data fusion, SKU large model research and development, as well as the capability building of retail full-scenario perception, digital human store managers, and data governance systems.
The "AI Store Clerk" adopts a dynamic operation model of "restocking during idle times, selling during busy times." During peak customer flow periods, it engages in greeting interactions, product recommendations, planned picking, self-service retrieval, transaction fulfillment, and repurchase incentives. During idle periods, it performs restocking, dynamic inventory counting, intelligent replenishment, and data analysis and operational decision-making. The entire system establishes a data flywheel of "real transactions—environmental data collection—model training—scenario deployment—continuous iteration," where offline transaction data continuously feeds back to enhance robot capabilities, achieving self-optimization of store operational efficiency and decision-making precision.
SenseMart is an ecosystem enterprise under SenseTime's "1+X" strategic ecosystem targeting the retail industry. It completed its Series A financing in January 2025, led by Jinshajiang Venture Capital with tens of millions of yuan, becoming the first "1+X" ecosystem enterprise to complete independent financing. SenseMart CEO Huang Di previously stated that for AI to truly land, it requires doing a lot of "dirty and tiring work" in vertical industries, from debugging SKU recognition accuracy to environmental adaptation for different retail scenarios, each requiring the R&D team to go deep into the front lines. Currently, the embodied intelligent retail track is undergoing a transition period from cash-burning R&D to running data models in real commercial scenarios. The implementation of the 100-store plan will accumulate larger-scale offline real transaction data and operational experience for SenseMart.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
