Daxiao Robotics and Kuaikeda have built an embodied retail skills training base in Deqing, Zhejiang, applying the company’s dual-brain architecture and embodied recursive self-improvement, or RSI, to instant retail for the first time.

Using a general model plus a Kaiwu world model, the two are running multi-robot collaborative learning and real-machine feedback loops around high-frequency tasks such as food preparation, bagging, shelf tidying and fine manipulation. The aim is to generate, refine and share retail skills that transfer between machines.
Three areas are in focus. In food preparation, robots handle macaron picking, oden serving, sandwich assembly, microwave heating of buns and fruit placement along a pick, process and plate chain. They work on fragile, soft and skewered ingredients, improving fine grasping, ordered operations and multi-layer assembly, and exploring spill prevention, oven door handling and button setting to move from single actions to full meal preparation for convenience store hot food zones.
In bagging and store tidying, the tasks cover packing, safety checks, hanging and retrieving goods, and arranging drinks and snacks. Mixed boxes, soft packaging, dense shelves and varying hook heights are used to strengthen recognition, classification and placement, with work extending to multi-item packing, sealing and labelling, expiry checks and replenishment.
In stationery and small item handling, robots work on transparent items, glasses cases, coffee and tea sachets, stamping and basket tidying. Reflective and soft objects in dense arrangements are used to improve grasping precision, pressure control and classification, with tasks expanding to stationery shops, coffee corners and checkouts.
The base runs dozens of robots learning at the same time, exploring how atomic actions and execution experience can be shared across machines so that what one robot learns serves the others. Under the plan, Daxiao and Kuaikeda will push thousands of retail robots into front-line stores in 2027.
Editor’s note: This is an adapted translation of the original Gasgoo report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://auto.gasgoo.com/news/202611/19I70478462C601.shtml.