Adaptive Edge Cloud technology behind Amazon Fresh could enable more contactless shopping


Amazon has opened its second till-free grocery store in the UK, just a couple weeks after its first launch in Ealing earlier this month. The move marks a significant change for the retail industry, as shops start to prepare to re-open after COVID-19 accelerated the shift towards more online sales.

Amazon’s cashless shop in London uses Amazon’s “Just Walk Out” technology, which uses sensor fusion to track what a person purchases. In-ceiling cameras, shelf level sensing, real-time image recognition and deep learning to automatically add items picked up from a shelf to that person’s store app. Those same technologies can also recognize when a product is removed from the customer’s physical shopping cart and returned to the shelf. But that’s a lot of instant data transfer between the store and the cloud– we’re talking near-petabytes of data transfer on a daily basis, along with a need for close-to-instant compute and processing of that data to deliver a seamless experience to the customer.

To make this happen, Amazon will be relying on cloud compute capabilities located either in or near the store, otherwise known as an Edge Cloud. Edge Cloud data centers such as these are set to become the norm. When combined with 5G rolling out across the planet, a resilient, faster, more immersive experience is created. The global edge data center market is expected to increase by more than $7 billion by 2024.

To enable these applications to run seamlessly, the network of edge clouds needs to be architected in a way that it can be flexible and affordable in the long run. These networks need to be adaptive and responsive and be able to readjust resources as needed to meet the demands of the application layer, and ensure they can manage, secure and automate the network end-to-end. If we enable a more programmable infrastructure that can adapt to support it, the way we shop in the future could become a blend of contact free and in-store shopping.