Amazon is working on a technology to replace barcodes

The American e-commerce giant processes an incessant flow of goods every day in its various warehouses around the world. To go faster, Amazon currently relies on the barcode technology used by their robotic arms to process the various products to be delivered.

Credit – Wilfried Pohnke/Pixabay

However, this option is far from giving optimal results. This situation has therefore led Amazon to consider other solutions that better suit the characteristics of their sorting robots. In a press release issued on Friday, Amazon, which does not only record bad news, indicated that it had found another technology that should better meet their expectations. It would be an overtrained algorithm using artificial intelligence, capable of optimizing the performance of its robots.

An AI-based system to make robotic arms more efficient

The e-commerce giant claims to have a computer system capable of studying items passing one by one on conveyor belts. According to information obtained on Friday, the computer tool developed with the help of an expert in artificial intelligence (AI) will above all make it possible to solve the current deficit of robots.

Indeed, the latter is still unable to correctly read the various barcodes affixed to the articles in order to sort them correctly. This leads to errors when shipping items to customers. Amazon’s goal is therefore to completely dispense with the use of barcodes in order to optimize its control chains. Nontas Antonakos, Applied Science Manager at Amazon’s Computer Vision Group in Berlin, said:

“Solving this problem so robots can pick up items and process them without having to find and scan a barcode is fundamental. This will help us get packages to customers faster and more accurately.”

Clearly, the solution to this problem takes the form of a system called “multimodal identification”. To get there, Amazon’s AI experts had to start by building a library of product images. The images as well as the data regarding the dimensions of the products fed the first versions of the algorithm. Subsequently, the use of cameras to continuously take new images of objects has served to train the model.

A testing phase and an accuracy rate of 99% according to Amazon

When it was first used, Amazon’s estimates of the algorithm’s accuracy rate were between 75 and 80 percent. Beginnings that the company then considered “promising”. In its press release, Amazon claims that the system’s accuracy rate now tops out at 99%. In fact, since the first problem that occurred during a Prime Day promotion, the algorithm has improved a lot. Thanks to some changes made by the development team, the vision identification system can assign confidence ratings to its evaluations.

But everything is not perfect. According to Amazon’s AI team, it will be difficult to refine the identification system to evaluate products handled by humans. That is why the ultimate goal is to have them manipulated directly by robots.

Before rolling out to all Amazon companies globally, the company is putting the computer vision algorithm through a testing phase at its facilities in Barcelona and Hamburg. The first benefits are already felt. In fact, Amazon claims that the multimodal identification system has helped speed up package processing times in these two cities. Once sufficiently established, we envision that the technology will be shared across all of the company’s facilities.

Source: CNET


Leave a Comment