正行创新 Unveils Physical AI for Retail: Robots Working the Shelves 24/7

On October 8, 2026, at the 22nd Asia-Pacific Retailers Exhibition (APRCE 2026), 正行创新 (Zhengxing Innovation in pinyin) announced the launch of a Physical AI solution for stores. The working formula: a robot performs routine operations on the sales floor, while a software layer hands it tasks and calls in a human when the machine can't cope. The company itself calls the product the world's first physical intelligence service for retail that runs 24/7. That is a marketing claim, and no one has independently verified it.

What exactly was shown

At the core of the solution is the "具身大脑", or "embodied brain". Chinese companies use the term for a model that links perception, planning and action for a robot in a real environment. The available description points to two components.

The first is hardware. It is a robot aimed at three practical tasks: restocking products on shelves, walking the store for inspection and carrying loads along the way. The set of tasks is narrow and down-to-earth. The company does not promise a general-purpose humanoid.

The second component is a unified software platform. It distributes tasks among machines, tracks their status and lets a human manually take over control. This layer is what turns a single robot into a service that can stay switched on around the clock.

• Event: APRCE 2026, October 8, 2026
• Developer: 正行创新
• Technology base: the "具身大脑" model
• Robot tasks: shelf restocking, sales-floor inspection, carrying
• Platform tasks: task assignment, monitoring, human takeover
• Claimed operating mode: 24/7

Why a store is a hard environment for a robot

A warehouse is predictable: racks stand in a grid and routes are fixed. A sales floor is different. Shoppers walk around, carts end up anywhere, products get moved every day, and promo displays appear and disappear on the marketing team's schedule. Chinese analysts call such spaces "open scenarios", which sets them apart from closed production zones.

That is why human takeover is the most important part of the announcement. It acknowledges that full autonomy in a store is not yet achievable. An operator who sees the machine's status and can take control turns a robot's errors into routine work episodes. Retail robots will still make mistakes, and the question is what each mistake costs.

The 24/7 mode also makes practical sense. Shelf restocking and night rounds can be moved to the hours when no people are on the floor, and in the morning the shopper sees a fully stocked display. That is, of course, if the robot manages without help.

What the announcement lacks

The available excerpt contains no figures that would allow the solution to be evaluated. It is unknown how many SKUs the robot handles per hour, how accurately it recognizes shelves or what payload it can carry. There is no data on price, on the sales model (purchase, lease or subscription), or on which chains already use the solution. It is not disclosed how many operators are needed per machine or how often manual takeover occurs. Without these metrics, the "world's first" claim remains a slogan.

How to approach such solutions in a pilot

A retailer considering a system like this should ask for several concrete things:

• the share of tasks completed without human intervention, measured on real shelves rather than a demo stand;
• the platform's response time to a failure and the average duration of a manual takeover;
• integration with inventory systems: the robot needs up-to-date planograms and stock data;
• requirements for store infrastructure: connectivity, charging, aisle width;
• the total cost per hour of operation, including operators, maintenance and downtime.

For now, 正行创新 has shown an architecture: a robot plus a dispatch layer plus a human in the loop. It is a sensible design for open spaces. Whether it becomes a working business will be shown by the first deployments and the data from them.