Shoppers buying loose apples, cucumbers or other produce usually have to find the correct item on a self-checkout screen before paying. A system being tested at three Rimi supermarkets in Vilnius, Lithuania’s capital, aims to shorten that process: a camera captures the product on the scales and artificial intelligence immediately suggests the most likely options.

The technology does not make the final decision. Shoppers must confirm the product on the screen, allowing them to check and correct the system’s suggestion when necessary.

The system uses Retellect AI technology. A camera records the product placed on the scales, while an algorithm assesses its colour, shape and other visible characteristics. The checkout then displays the most likely matches instead of requiring the shopper to search through a long list of fruit and vegetables.

Andrius Kalašinskas, a representative of retail technology company StrongPoint, which installed the system, said the shopper’s choice remains decisive. The technology is intended to reduce the time spent searching for products that may look similar on a checkout display.

The algorithm has so far been trained to recognise about 30 of the most frequently purchased weighed fruits and vegetables. These are products that customers commonly need to identify manually at self-checkouts.

According to estimates provided by the company, the system can save an average of about 10 seconds for each product entered. When several weighed items are purchased, the accumulated saving could shorten the overall shopping trip by several minutes.

The difference is likely to be most noticeable when a checkout presents numerous similar options. Shoppers do not need to know the exact product name immediately or spend as long comparing small images. The system narrows the selection without preventing the customer from choosing another option.

The technology is also designed to reduce product-identification errors and the associated risk of losses for retailers. For store employees, it could mean fewer routine corrections at self-checkouts and more time to assist customers with other requests.

During the trial, the system recognises the most frequently purchased weighed products with reported accuracy of more than 97%, according to a statement prepared by communications agency Publicum and published by Lithuanian news agency ELTA. This is a performance figure from an ongoing test, not a guarantee that every item will be identified correctly.

Recognition can be affected by supermarket lighting, variations in a product’s shape and colour, and the way it is positioned on the scales. The algorithm must also assess produce inside bags, which may conceal part of an item or alter how its colour appears to the camera.

Distinguishing between similar apple varieties is among the most difficult tasks. The system must detect subtle differences in colour, pattern and shape that may also be difficult for shoppers to assess. Human confirmation therefore remains the main error-control measure: the AI proposes possible matches, but the customer selects the final product.

The technology is currently being tested only at three Rimi stores in Vilnius. The statement did not identify their locations, and the trial should not be interpreted as a service already available throughout the supermarket chain.

Vaidas Lukoševičius, chief executive of Rimi Lietuva, said the trial is intended to make shopping faster while reducing the volume of routine work handled by employees. Any further expansion will depend on results collected in working supermarket conditions.

The next planned stage is to train the system to recognise different types of rolls, bread and other baked goods. Wider trials are also planned in Latvia, another Baltic country, later in 2026.

Context & actions
About this article

Report
Share

Source check
Šaltinio pėdsakas

Publikacijoje bandymo mastas, tikslumo rodiklis ir pirkėjo vaidmuo atskirti nuo dar tik planuojamų technologijos galimybių.

Checks

Patikrinta, kad sistema bandoma trijose „Rimi“ parduotuvėse Vilniuje.
97 proc. rodiklis pateiktas kaip bandymo metu pasiektas atpažinimo tikslumas.
Patvirtinta, kad galutinį prekės pasirinkimą atlieka pirkėjas.
Kepinių atpažinimas ir bandymai Latvijoje nurodyti kaip tolesni planai.

SourceELTA

ScopeVilnius

Updated2026-07-21 11:43

Report

Article contextPeople & topics2#3Andrius KalašinskasVaidas Lukoševičius