How Amazon Go Works

What Amazon Go Is

Amazon Go is a line of convenience stores built around a single idea: no cash register, no scanning, no line. A shopper opens the Amazon app, generates a code, and taps it at an entry gate (some locations instead use a palm scan through Amazon One). From that point, the store identifies which account is shopping and starts tracking that person's activity until they leave.

The first Amazon Go store opened to the public in Seattle in January 2018, after roughly a year of employee-only testing. It sits in the same family as Amazon Fresh stores, which have experimented with several checkout formats over time, including smart carts. The lineup of open stores and the technology used inside them has kept changing since launch, so check current store listings before assuming a specific location works a particular way.

The Sensor and Camera Layer

The physical infrastructure is a dense sensor network rather than a single gadget. Ceiling-mounted cameras cover the shelves from multiple angles, and weight sensors sit under the products on each shelf. There are no RFID tags on individual items; products stay exactly as a store normally stocks them.

This means every shelf interaction produces two kinds of raw signal at once: a visual stream showing a hand or body near a shelf, and a numeric signal showing a weight change. Neither signal alone is reliable. A camera can lose a hand behind another shopper; a weight sensor can't tell you which of several nearby items moved. The system's job is to reconcile the two.

How the System Tracks a Shopping Session

This reconciliation is a sensor fusion problem, the same general class of problem used in robotics and autonomous vehicles. Computer vision models track each person's position and pose as they move through the store. Separate models watch shelves for hand entry and exit. When a weight sensor reports a change, the system correlates it with the nearest tracked person and the nearest camera view to decide what was picked up or put back.

Each confirmed pickup or return becomes an event tied to that shopper's session. The store does not need to know who a person is by face; it only needs to keep a consistent link between a tracked body and the account that scanned in at the gate. Amazon has stated that this tracking does not rely on facial recognition.

A Simplified Model of the Virtual Cart

Conceptually, the running total of what a shopper is holding, often called a virtual cart, can be modeled as an event log that gets reduced into a current state. This is a simplified illustration of the idea, not Amazon's actual implementation:

cart = {}

def handle_event(event):
    # event: shelf_id, item_id, weight_delta, timestamp
    if event.weight_delta  0:
        cart[event.item_id] = max(0, cart.get(event.item_id, 0) - 1)

def checkout(cart):
    return {item: qty for item, qty in cart.items() if qty > 0}

Each shelf event mutates the cart; nothing is finalized until the shopper leaves, at which point the current state of the cart becomes the order. This event-driven pattern, appending raw signals and deriving state afterward, is the same approach used in event sourcing for transactional systems, applied here to physical shelf activity instead of database writes.

Just Walk Out as a Licensed Platform

The technology behind Amazon Go is distinct from the Amazon Go brand itself. Amazon packaged the camera-and-sensor system as a separate product called Just Walk Out and licensed it to third-party retailers, stadiums, airports, and other venues that wanted checkout-free sections without building their own computer vision stack.

Store formats have also diverged over time, and that mix keeps changing. Some Amazon Fresh grocery stores originally used full Just Walk Out tracking; Amazon has since moved toward Dash Cart, a smart shopping cart with its own scanner and screen, as the primary checkout-free option in Fresh stores, and has said it was pulling Just Walk Out back out of that format. Just Walk Out remains in use at Amazon Go stores and at third-party licensees. Because this keeps shifting, check justwalkout.com or aboutamazon.com for what is currently deployed where.

Trade-offs and Limitations

• Installation cost and calibration: covering every shelf angle with cameras and wiring every shelf with weight sensors is capital-intensive and has to be re-tuned whenever a store's layout or product mix changes.
• Accuracy under crowding: dense foot traffic, items that look similar, or shelves stocked unevenly all make the vision and weight signals harder to reconcile correctly.
• Scaling to large formats: the approach was proven in small-footprint convenience stores; extending it cleanly to a full grocery store with thousands of SKUs is a materially harder tracking problem, which is part of why larger Fresh stores have leaned on Dash Cart instead.
• Connectivity and compute: the system depends on continuous, low-latency processing of many video streams plus sensor data, which raises infrastructure requirements compared with a normal point-of-sale setup.
• Privacy concerns: even without facial recognition for billing, continuous video tracking of shoppers' movement raises questions that vary by jurisdiction and that retailers adopting the platform need to address directly with customers.

Common Mistakes and Misconceptions

A few points get confused often enough to call out directly:

• It is not RFID-based. No tags are attached to products; tracking comes entirely from cameras and weight sensors.
• Amazon Go, Just Walk Out, and Dash Cart are not the same thing. Amazon Go is a store brand; Just Walk Out is the licensable tracking technology; Dash Cart is a separate smart-cart product used in some larger stores.
• Not every location works identically, and that has changed over time. Entry method (app code versus palm scan) and checkout method (full Just Walk Out tracking versus Dash Cart) can differ by store, and Amazon has shifted some Fresh locations away from full camera-based tracking toward Dash Cart. Assume nothing about a specific location without checking current signage or Amazon's own materials.