Robots may be the future, but robotic arms are apparently no good at using the good old barcode. Barcodes can be hard to find and can be affixed to oddly shaped products, which robots can’t troubleshoot very well.
As a result, Amazon said on Friday it had a plan to kill the barcode.
Using images of items in Amazon’s warehouses to train a computer model, the e-commerce giant has developed a camera system that can monitor items moving along conveyor belts one by one to ensure they are correspond to their images. Eventually, AI experts and roboticists at Amazon want to combine the technology with robots that identify objects while picking them up and turning them over.
“Solving this problem, so that robots can pick up objects and process them without needing to find and scan a barcode, is fundamental,” said Nontas Antonakos, head of applied science at the computer vision group. from Amazon in Berlin. “This will help us deliver packages to customers faster and more accurately.”
The system, called Multimodal ID, won’t completely replace barcodes anytime soon. Products in Amazon’s warehouses will need to have barcodes as long as the outside companies that make and ship them rely on technology to identify and track inventory. Amazon’s new system is currently in use at facilities in Barcelona, Spain, and Hamburg, Germany, the company said, adding that it is already speeding up the time it takes to process packages there. The technology will be shared among Amazon’s businesses, so it’s possible you could one day see a version of it at a Whole Foods or other Amazon-owned chain with in-person stores.
Amazon has integrated computer vision into other products. You can ask an Echo Show smart display, “Alexa, what do I want?to get help recognizing objects around the house. The feature is called Show and Tell and was designed for the visually impaired. Smartphone manufacturers and social media companies also included AI functionality in camera and photo apps, automatically categorizing photos, for example.
The problem the system eliminates — incorrect items coming on the line to be sent to customers — doesn’t happen too often, Amazon says. But even infrequent errors add up to significant slowdowns when you consider the number of items a single warehouse processes in a day.
Amazon’s AI experts had to start by building a library of product images, which the company had no reason to build before this project. The images themselves as well as data on the dimensions of the products fed the first versions of the algorithm, and the cameras continually capture new images of items with which to train the model.
The algorithm’s accuracy rate was between 75% and 80% when it was first used, which Amazon considered a promising start. The company claims the accuracy is now 99%. The system encountered an initial problem when it failed to detect color differences. During a Prime Day promotion, the system could not distinguish between two different colors of Echo Dots. The only difference between the packages was a small blue or gray dot. With some retooling, the ID system can now assign confidence scores to its assessments that flag only items it is certain are incorrect.
Amazon’s AI team says it will be difficult to fine-tune the multi-modal ID system to assess products handled by people, so the ultimate goal is to have robots manipulate them instead.