Summarize this content to 2000 words in 6 paragraphs
Amazon says a new multimodal AI foundation model improves its “Just Walk Out” system. (Amazon Photo)
Amazon has developed a new AI model for its “Just Walk Out” technology to streamline the process of tracking shoppers’ activity and generating receipts, helping to accelerate the rollout of the checkout-free approach to more third-party stores.
The new multimodal AI foundation model analyzes different types of sensor data simultaneously rather than sequentially as the prior approach did, improving its ability to automatically identify items that shoppers take from store shelves.
Amazon’s new approach “allows us to generate receipts faster, and more accurately, more efficiently,” said Jon Jenkins, vice president of Amazon’s Just Walk Out technology, during a briefing for reporters this week at the Amazon campus in Seattle, where the company demonstrated how the new AI model works.
The new model is slated to roll out to existing stores over the next month.
Amazon says the new system uses the same type of transformer-based machine learning models used by generative AI applications, although it has been in development for more than two years, prior to the debut of OpenAI’s ChatGPT.
Amazon says the new multimodal AI model improves the accuracy and efficiency of its Just Walk Out checkout-free technology. (Amazon Image)
Amazon also detailed the new system in a blog post this morning.
Just Walk Out is used by some of Amazon’s first-party retail locations, such as the Amazon Go convenience stores, even after Amazon said that it would remove Just Walk Out from its large-format Amazon Fresh stores.
The Just Walk Out technology team operates independently from the physical stores team as part of the Amazon Web Services cloud division, and offers the checkout-free technology to third-party locations.
It’s known inside Amazon as JWO, which some pronounce “jay-woah.”
Amazon said in April that it would double the number of third-party Just Walk Out stores this year, adding more new locations than in any previous year.
Just Walk Out is now available in 170 third-party locations in the U.S., the UK, Australia, and Canada, including airports, stadiums, universities, and hospitals.
The company previously debunked reports that humans are watching through cameras to tally up every shopper’s bill. People have been used to make annotations as part of the training process for the AI that powers Just Walk Out, following standard industry practice, but the technology itself works autonomously.
Jenkins said there are five main use cases where Just Walk Out makes sense for third-party retailers: reducing lines, enabling 24/7 operations, reducing theft, optimizing labor, and enabling more efficient retail space in new venues.
Just Walk Out uses a combination of cameras and (for smaller items) weight sensors in shelves, in conjunction with a catalog of items in the store. The system also uses three-dimensional interactive maps of stores that can now be generated with the built-in LiDAR in the cameras in Apple iPhones or iPads.
During demonstrations for reporters, Amazon showed complicated situations that can pose challenges for the system, such as customers grabbing two items at once, or the glass on a refrigerated case fogging up and blocking the view of an item. The new AI system is more adept at dealing with these challenges, the company said.
Amazon has separately developed a RFID-enabled version of the technology that is available at several stadiums in the U.S. and internationally, and is slated to be rolled out to more venues in the months ahead.
