21 Nov 2019

It’s a Friday evening and you’ve decided to cook fish tacos. So you pop into your local supermarket on the way back from work. It takes several minutes just to hunt down the cilantro. Or, more accurately, to locate the empty shelf where the cilantro should have been. After another 10 minutes waiting in line for the single-staffed checkout lane, you finally get home to realize you forgot to buy tortillas. And that’s how you end up spending your Friday night eating leftover casserole.

Data moves at almost the speed of light, but groceries don’t. As the spread of high-speed internet renders information transmission ever-faster, the efficiency of the necessary physical transactions involved in buying and selling goods has lagged behind. That’s about to change.

“A lot of companies both big and small are working on using sensor technology and machine learning to improve the shopping experience ,” said Gustavo Martinez, a system engineer at our company. “Customers are frustrated by things like standing in a long checkout line, or finding out that they don’t have the item they want, or that it’s more expensive than somewhere else.”

A personal shopper in your pocket

The combination of machine learning and GPS technology already allows retailers to deliver personalized advertisements as a potential customer enters their vicinity. The next step is the use of in-store sensors, such as Bluetooth beacons, to deliver hyperlocal promotions at the level of the individual shelf.

These might trigger a custom notification on a smartphone—such as a half-price offer on vanilla wafers to the customer who will spend several minutes staring at the cookie aisle. Alternatively, replacing paper price tags with LCD displays will enable flexible offers to be displayed on the shelf itself, changing as different customers are approaching.

These smart displays can also guide a customer around a store, Gustavo said. “The store’s app can plot out the most efficient route to pick up all the items in your list, and we can have the in-shelf displays light up as you approach to make it easy to locate the item that you’re looking for."

The end of the line for the checkout lane

Among the most significant changes to the in-store experience has been the rise of self-checkouts. These aren’t just about saving staffing costs for stores.

“The main thing is getting rid of the need to stand in line to check out,” Gustavo said. “At least in my case, having to wait that additional 10 or 15 minutes is my least favorite part of going to a store.”

Self-checkouts aren’t perfect, however. There’s still a relatively laborious process for entering uncoded items, such as loose fruit, and the need for a store assistant to dart between sale points to assist with problems and age-restricted items.

“Some companies are looking into integrating cameras into the self-checkouts that can use machine vision to identify the items you’re buying,” said Aldwin Delcour, a systems engineer at our company. “Instead of having to search through a whole set of menus, you can just put your apple in front of a camera and the system can automatically identify it.”

While more numerous self-checkouts haven’t eliminated the line altogether, the end of the line may be coming. At stores on the cutting edge of retail automation, customers scan their phone as they walk in, and a combination of cameras and in-shelf sensors tallies up the items put into their basket and automatically bills them when they leave.

Currently, this requires sending streams of data from potentially hundreds of thousands of stores up to the cloud for processing by machine learning algorithms.

“That’s an enormous amount of data being siphoned off, which can present significant challenges,” Gustavo said. “So we’re looking at how that data can be processed in the store itself to reduce that load.”

TI mmWave sensors, which bounce high-frequency radio waves off an object to precisely identify its shape, size and distance, can simplify the recognition task, potentially allowing it to be performed in-store on our Sitara™ processors, specifically designed for low-power machine learning applications.

The highly-sensing store that’s never out of stock

Smoothing a customer’s journey through a store also includes making sure items they want are where they should be. Ubiquitous sensing will not only enable stores to track customers but also stock, ensuring that low-levels of an item can be detected instantly and supply ordered.

"A store might have a spring mechanism so that when you take an item, a new one is pushed forward," Aldwin said. “You can put a sensor in the back that detects how far it has moved, and then gives a signal to a centralized computer that the inventory is low and it might be time to order the next shipment.”

Once the inventory order is placed, the same technology that guides customers around a store can also be used to guide stock pickers around a warehouse, making the process of filling an order much faster and more efficient.

A Future Friday evening

The future of grocery shopping could look like this: It’s a Friday evening, and the app for your local supermarket sends you a recipe for fish tacos. Based on your previous shopping behavior, the company’s machine learning algorithms have built up a profile of you as a Mexican-food lover who enjoys Friday night cooking, and the recommendation is perfect. You click to add the ingredients to a digital shopping list and head to the store.

As you walk through the door, a notification pops up offering a map of where all your ingredients are located. The label beneath the relevant item lights up as you approach and nothing is out of stock.

Once all the items are in your basket, you walk straight out the door. No security guard chases after you. Instead, your phone delivers a receipt and informs you that all of the items have been charged to your account.

The whole process took a few minutes, and you arrive home early with a full set of fish taco ingredients. The rest is up to you.