Asian woman and surgical mask shopping some food in supermarket

Shopping habits of in-store customers analysed by floor-walking robot

Image credit: Thanakorn Hormniam/Dreamstime

Researchers in Russia have developed a novel method for customer behaviour analytics and demand distribution, taking inspiration from radio frequency identification (RFID) stocktaking.

Autonomous robotic systems that already pervade our daily lives face a host of challenging tasks, including stocktaking in a rapidly changing environment. To tackle this, a team at Skoltech’s Intelligent Space Robotics Lab in Moscow has proposed a novel method that helps build models depicting location-related demand dependencies and precise locations of lost and moved items.

The team, led by Professor Dzmitry Tsetserukou from Skoltech Space Centre, developed a robot capable of reading RFID tags with an accuracy of 0.3m. The robot monitors shoppers, notes the locations they find the most attractive and predicts demand. As a result, it gives useful tips to the retailer on where it is best to place an item in order to increase sales and profits.

“Existing solutions lack applicability to real-life situations in retailing, which may cause an unforeseeable loss of sales. Our solution provides the retailer with exhaustive information about demand distribution by using a mobile robot for autonomous stocktaking at RFID-equipped stores. Our research stands apart from other related works because of the mere size of the underlying set of real-world data we collected over 10 months,” says co-developer, Alexander Petrovskiy.

A Skoltech robot analyses shoppers' behaviour

Image credit: Skoltech

Professor Tsetserukou explained that the Skoltech team has developed 'Michelle', an RFID-capable autonomous robot, for Decathlon stores. The researchers added the robot helps considerably reduce the number of RFID-reading errors caused by the human factor and makes the inventory faster and cheaper. 

“We came up with an idea to benefit from 'Big Data' collected over 10 months of the robot’s operation at the store. Specifically, we wanted to estimate the tag density change over the entire purchase area,” Tsetserukou said.

The team proposed a probabilistic model for estimating tag locations with an accuracy of 0.3m, according to the researchers. They then constructed the hit map of the tag density dynamics, which showed the maximum and minimum purchase quantity areas.

Tsetserukou concluded such findings are very important for retailers in relocating the goods in order to maximise sales profits and predict seasonal demand profiles.

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles