The development of Amazon’s new AI-based Just Walk Out (JWO) technology represents a significant advancement in frictionless shopping experiences, leveraging multi-modal foundation models and transformer-based machine learning to accurately track customer interactions and purchases.
Key Takeaways:
- Amazon’s upgraded JWO system uses AI to analyze data from various sensors, improving accuracy in complex shopping scenarios and making the technology easier for retailers to deploy.
- The system’s self-learning capabilities reduce the need for manual retraining, allowing it to adapt to store layout changes and accurately identify misplaced items.
- JWO’s integration of RFID technology offers a more cost-effective and less complex solution for retailers, potentially expanding its application to temporary retail settings.
How JWO Works: The process of building a JWO-enabled store involves creating a 3D map of the space, dividing it into product areas called “polygons,” and installing custom cameras and weight sensors:
- JWO tracks the orientation of the head, left hand, and right hand to detect when a user interacts with a polygon.
- By fusing inputs from multiple cameras and weight sensors with object recognition, the models accurately predict whether a specific item was retained by the shopper.
- The improved AI model can now handle complex scenarios, such as multiple shoppers interacting with products simultaneously or obstructed camera views.
The Role of Edge Computing: JWO’s requirement to process and fuse information from multiple sensors in real-time highlights the importance of edge computing for real-world AI inference use cases:
- All model inference is performed on computing hardware installed on-premise, which is fully managed by Amazon and priced into the total cost of the solution.
- The data generated by JWO is too large to stream back to inference models hosted in the cloud, making edge computing a critical layer for these applications.
Scaling Up with RFID: Amazon is rapidly integrating RFID technology into JWO, simplifying the infrastructure requirements and potentially expanding its application to temporary retail settings:
- The AI architecture remains the same, featuring a multi-modal transformer fusing sensor inputs, but without the complexity of multiple cameras and weight sensors.
- Many retail clothing items already come with RFID tags from the manufacturer, making it easier for retailers to implement this flavor of JWO.
Broader Implications: The development of JWO demonstrates the high-risk nature of R&D in enterprise AI, IoT, and complex technology integration, as well as the potential for these investments to transform the retail industry:
- Large dollar hard-tech AI investments only make sense for companies like Amazon, which have the resources and scale to justify the risk and potential rewards.
- The successful implementation of JWO could lead to a wider adoption of frictionless shopping experiences, changing consumer expectations and forcing other retailers to adapt to remain competitive.
Inside Amazon’s new ‘Just Walk Out’: AI transformers meets edge computing