Dixy tests robot for shelf prices in one St Petersburg store

By Ekaterina Larina

Dixy tested a shelf-price robot in one St Petersburg store, taking a mobile unit from autonomous movement through image capture to a final recognition report. The pilot demonstrates a complete technical workflow, but it does not establish performance across Dixy’s network.

Dixy, part of the Magnit group, and Tevian ran the trial. The robot originally served as a cleaning machine; the partners adapted it to travel along retail shelving and equipped it with a separate camera module. It photographed shelves while moving, after which a computer-vision system recognized price tags and compared them with the store’s product matrix.

“Dixy will continue testing robots for recognizing price tags,” the retailer said. The announcement gave no date, store target, accuracy threshold or budget for that promise.

How did the recognition system work?

The software used fuzzy search, which tolerates abbreviations and differences in spelling when matching label text with matrix entries. The technology looks for a close textual match without requiring identical character sequences. Yet the pilot disclosed no figures for tags read, false matches, missed tags or performance against human inspectors.

What did the pilot prove?

Testing took place in one Dixy store in St Petersburg. The partners checked the full scenario, from the robot’s autonomous pass to a report on recognized price tags. Their stated conclusion was that a mobile robotic platform can supply images for automated trade-space auditing.

A one-store result is not like-for-like evidence of chainwide audit performance. The trial disclosed no purchase or maintenance cost, operating time, labor savings or payback period. Without those numbers, the commercial case remains unquantified.

What changes in the next tests?

Dixy plans to widen coverage of the sales floor and process more price tags, then connect the output with internal analytics systems. The company set no schedule, store target, accuracy goal or integration deadline. Broader coverage also creates more opportunities for false matches, so the pilot cannot support a forecast of results at scale.

Integration remains unspecified. The announcement does not say which internal users would receive alerts or how results would enter existing workflows. That leaves the operating benefit unproven even if image capture and text matching function as intended.

What else might the robots do?

Dixy sees potential uses beyond price checks:

• Checking whether products remain in stock.
• Recognizing stock-keeping units, or SKUs.
• Analyzing product display and comparing it with the store planogram.
• Handling other sales-floor tasks.

These remain possible applications rather than tested results. The pilot focused on price-tag images, and the company did not say when broader functions would enter testing.

The broader tasks would also require separate validation. A stock check must distinguish an absent product from an unrecognized image, while a display check needs the correct product and planogram mapping. The pilot supplied no measured results for either task.

The announcement also leaves open whether robots would operate around customers, blocked aisles or active store schedules. Those questions matter for a retail deployment, but the pilot did not answer them.

The pilot shows that Dixy and Tevian assembled a working end-to-end process. It does not show that robots can replace manual audits at scale; the available evidence supports the workflow, not the business case.