Amazon‘s cameras can distinguish goods using computer vision, eliminating the need for barcodes. Eventually, robots will be supported by the system. Amazon’s Strategy to Remove Barcodes
Although robots may be the technology of the future, robotic arms seem incapable of employing a time-tested technique: the barcode. In addition, barcodes may be difficult to locate and can be applied to irregularly shaped objects, both of which are issues that robots struggle to address.
Consequently, the firm said on Friday that it intends to eliminate barcodes.
Using photos of things in Amazon warehouses to train a computer model, the e-commerce behemoth has devised a camera system to monitor individual objects moving down conveyor belts to ensure that they match their representations. Eventually, Amazon’s AI researchers and roboticists want to integrate the technology with robots that can detect goods while picking them up and rotating them.
“Solving this challenge so robots can pick up and process products without having to locate and scan a barcode is crucial,” said Nontas Antonakos, an applied science manager at Amazon’s computer vision department in Berlin. It will help us deliver items to clients faster and more precisely.
The multi-modal identification system will not completely replace barcodes just yet. According to Amazon, it is now used at facilities in Barcelona, Spain, and Hamburg, Germany. Nonetheless, the corporation claims it has already reduced the time required to handle parcels there. In addition, the technology will be shared across Amazon’s companies, so it’s feasible to see a version of it at Whole Foods or other Amazon-owned brick-and-mortar shops in the future.
Amazon has used computer vision on other items. For example, you may query an Echo Show smart display, “Alexa, what am I holding?” for assistance identifying household items. In addition, the Show and Tell function was created with visually challenged individuals in mind. Smartphone manufacturers and social media businesses have also included AI functions in camera and picture applications, such as the automated categorization of photographs.
Amazon states that the issue that the method avoids — faulty products being shipped to consumers — does not occur often. Considering the number of products a single warehouse handles daily, however, even uncommon errors might result in considerable delays.
Before this effort, Amazon did not need to establish a library of product photos, so the company’s AI specialists had to generate one from scratch. The photos and data on the goods’ dimensions fed the first versions of the algorithm, and the cameras continuously recorded new photographs of objects to train the model.
Amazon thought the algorithm’s first accuracy rate between 75% and 80% was a good start. The business claims that the current accuracy is 99%. Initial difficulties arose when the system failed to detect color changes. During a Prime Day offer, the system could not differentiate between two distinct Echo Dot hues. A little blue or grey dot was the sole distinguishing characteristic between the containers. With some retooling, the identification system can now give confidence scores to its ratings so that it only flags objects for which it is certain are erroneous.
It will be difficult for Amazon’s AI team to fine-tune the multi-modal identification system to evaluate things being handled by humans. Therefore, the ultimate objective is to have robots handle them instead. Amazon’s Strategy to Remove Barcodes
