Amazon’s newer Just Walk Out is a hybrid retail platform, not a single camera or chatbot. It combines computer vision, shelf and weight sensors, machine-learning models, a transformer-based multimodal foundation model, optional RFID, in-store edge computing, AWS cloud services and payment systems. The technology is improving, even as Amazon has narrowed its use in large grocery stores and continues selling it mainly to third-party retailers.
What changed in Just Walk Out
The customer promise remains simple: authenticate at the entrance, take products, leave and receive a receipt charged to the associated payment method. Depending on the deployment, entry can use a credit or debit card, mobile wallet, retailer app or QR code, employee badge, or Amazon One. Some installations support payment at the exit instead. The available method is determined by the retailer and location (Just Walk Out: How it works).
Three developments are often incorrectly treated as one:
- The original checkout-free system built a virtual cart from shopper, product and sensor events.
- Amazon’s July 31, 2024 upgrade added a multimodal foundation model using transformer-based machine-learning techniques to analyze camera and sensor information together (Amazon’s announcement).
- The current operating strategy emphasizes edge processing, additional sensors and action-recognition algorithms, while targeting third-party small-format stores and venues rather than every grocery trip.
Amazon has not published the model’s exact architecture, parameter count, training corpus, inference framework, edge chips or end-to-end latency. The production system should therefore be understood as a hybrid of models, sensors, rules, retailer integrations and operational controls—not as one publicly specified transformer doing everything.
#1 Best Overall
- The NVIDIA Jetson AGX Orin 64GB Developer Kit makes it easy to get started with Jetson Orin. Compact size, lots of connectors, and up to 275 TOPS of AI performance make this developer kit perfect for prototyping advanced AI-powered robots and other autonomous machines.
- The developer kit includes a Jetson AGX Orin 64GB module, and can emulate all the Jetson Orin modules. It supports multiple concurrent AI application pipelines with the NVIDIA Ampere GPU architecture, next-generation deep learning and vision accelerators, high-speed IO and fast memory bandwidth. Now you can develop solutions using your largest and most complex AI models to solve problems such as natural language understanding, 3D perception, and multi-sensor fusion.
- Jetson runs the NVIDIA AI software stack, and use-case specific application frameworks are available, including Isaac for robotics, DeepStream for vision AI, and Riva for conversational AI. You can save significant time with NVIDIA Omniverse Replicator for synthetic data generation (SDG), and by using NVIDIA TAO toolkit to fine-tune pretrained AI models from the NGC catalog.
- Jetson ecosystem partners offer additional AI and system software, developer tools, and custom software development. They can also help with cameras and other sensors, as well as carrier boards and design services for your product.
- With the computing capability of more than 8 Jetson AGX Xavier systems in a developer kit that integrates the latest NVIDIA GPU technology with the world’s most advanced deep learning software stack, you’ll have the flexibility to create tomorrow’s AI solution as well as today’s.
The real engineering problem: “Who took what?”
Recognizing a sandwich is relatively easy. Correctly charging the right sandwich to the right person after several shoppers move around it is the hard part.
Just Walk Out must reconstruct a transaction from physical events:
- Which person entered and which account or payment credential is associated with that person or group?
- Which shopper reached toward a product?
- Did the product actually leave the shelf, or was it only lifted and replaced?
- Was it placed in a bag, pocket, stroller or basket?
- Was it handed to another shopper or returned to a different location?
- How many identical units were taken?
- Did another shopper, a restock worker or an obstruction block the camera?
- Which items ultimately crossed the exit boundary?
Amazon describes using cameras, weight sensors and AI to determine the variety and quantity of selected items (Amazon’s laboratory description). This is a multi-person identity-association and temporal-event-reconstruction problem, not merely image classification.
Where transformer-based AI fits
Older-style computer-vision systems commonly divide the work into stages: detect objects, identify hands or bodies, estimate movement, then connect events with rules or separate models. Amazon says its newer multimodal foundation model can process camera and sensor information simultaneously rather than treating every pickup and put-back as an isolated linear sequence.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →In practical terms, a model may learn relationships across a sequence such as:
Rank #2
- 30-in-1 No-Solder Sensor Board, Plug and Play: Integrates 30 functional sensors including temperature & humidity, ultrasonic ranging, gas and motion sensors. Innovative common board design requires no soldering or complex wiring, and comes with a full set of accessories like 128G SD card, adapter board and acrylic mounting plates for zero-threshold experiments
- 8MP Gimbal Camera & Dual Servos for Professional Visual AI: The Starter Kit is equipped with an IMX219 8MP monocular camera and a dual-servo gimbal, supporting face and target tracking, and is ideal for AI edge computing scenarios such as intelligent monitoring, robot navigation, and automated recognition
- 38 Step-by-Step Python Tutorials, From Beginner to Practical Application: The Jetson Orin Nano Starter Kit comes with 38 well-designed Python tutorials progressing from basic programming to vision practice, covering all key knowledge of sensor control, embedded development and AI visual recognition for both beginners and advanced learners
- 11.6-inch IPS HD Screen & AI Voice Interaction System: Built-in 1366*768 resolution IPS screen eliminates the need for an external monitor, enabling one-device experimentation and visual feedback. The exclusive AI voice interaction system supports intelligent Q&A and voice command control for natural human-computer dialogue
- Rich Expansion Interfaces & Portable All-in-One Design: Features 2x I2C, 1x UART and 2 IO expansion interfaces to meet personalized experiment expansion needs; a custom carrying case integrates all components (11.81×7.87×3.94 inch), allowing AI experiments and demonstrations anytime and anywhere
- A hand approaches a shelf.
- A shelf or weight signal changes while the product disappears from view.
- The shopper turns and the item appears inside a bag.
- The product is not returned before the shopper exits.
A multimodal model can also be trained to generalize when a store is remerchandised, an item is misplaced, a new product is introduced or customer behavior differs from training examples. Amazon says this is the purpose of the 2024 upgrade (Amazon’s technical announcement).
That does not mean transformers replaced every previous component. A real deployment still needs product catalogs, tracking logic, sensor calibration, payment authorization, inventory and receipt services, exception workflows and staff controls.
Why edge computing matters
A checkout-free store produces continuous video and sensor events. Sending every raw camera stream to a distant data center would increase bandwidth cost, latency, connectivity dependence and the consequences of an outage. Stores may also be geographically far from AWS regions.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Amazon and AWS describe in-store hardware and microservices working with cloud services, and Amazon’s later strategy update refers to building edge compute to improve latency, accuracy, receipts and action recognition (AWS technical description; Amazon strategy update).
A defensible high-level architecture looks like this:
Rank #3
- Stability: Can be used stably for a long time
- Design: Robust design, easy to maintain
- Easy to install: simple operation, easy to install
- Application Scenario:Widely used in many industrial environments
- Correct use:Correct use can extend the service life of the product
Cameras / shelf sensors / RFID
↓
In-store edge systems
event filtering + local inference
↓
Multimodal AI pipeline
↓
shopper identity ↔ virtual cart
↓
cloud services / retailer systems
↓
receipt + payment authorization
The public descriptions do not establish exactly which inferences run locally, which run in the cloud, what hardware is installed, or how a store reconciles carts when connectivity fails. Retailers should require those details contractually rather than assume every deployment has the same topology.
The sensor stack behind the experience
Cameras
Cameras observe body and hand movement, shopper trajectories, product visibility and interactions with shelves. They provide the visual context needed to associate an item movement with a person, but they can be affected by occlusion, dirty or misaligned lenses, crowding and lookalike packaging. Amazon’s overview explains the camera-based workflow (How Just Walk Out works).
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsShelf and weight sensors
Weight or other shelf sensors provide nonvisual evidence: a local weight change, an approximate interaction location or confirmation that an item moved. They are especially useful when a hand or product is temporarily hidden. They can still be disturbed by shelf vibration, a shopper leaning on a fixture, a bag or a restocking action. AWS describes this sensor fusion approach (AWS).
RFID
RFID can identify tagged items directly and is useful for apparel, shoes, fan gear and other merchandise. Amazon has described RFID-enabled deployments for clothing and related products (Amazon’s RFID explanation). The trade-off is operational: tags must be applied, readers must cover the relevant area and receiving, inventory and tag-management processes must remain accurate. Vision can work with ordinary packaged goods but must handle blocked views and similar packaging.
Identity, payment and store systems
Entry credentials connect a shopper or group to a virtual cart. The completed cart then has to connect to payment authorization, tax, promotions, loyalty, inventory, refunds and the retailer’s point-of-sale systems. Just Walk Out can operate independently of Amazon One; Amazon says it does not require biometric information about the shopper (Amazon FAQ article). That does not mean there is no personal data: a payment method, entry event, receipt and account association remain sensitive transaction information.
Rank #4
- Brilliant AI Performance for production: on-device processing with up to 70 TOPS AI performance with low power and low latency, Due to the high thermal demands of Super mode, only the J30 Series supports upgrading to Super mode via the JetPack 6.2 update.
- Hand-size edge AI device: compact size at 130mm x120mm x 58.5mm, includes NVIDIA Jetson Orin NX production module, a heatsink, enclosure, and a power adapter. Support desktop, wall mount, fit in anywhere
- Expandable with rich I/Os: 4x USB3.2, HDMI 2.1, 2xCSI, 1xRJ45 for GbE, M.2 Key E, M.2 Key M, CAN and GPIO
- Accelerate solution to market: pre-installed JetPack with NVIDIA JetPack 5.1.1 on the included 128GB NVMe SSD, Linux OS BSP, 128GB SSD, support Jetson software and leading AI frameworks and software platforms
- Comprehensive certificates: FCC, CE, RoHS, UKCA
Human review: what is known and what is not
In 2024, reports characterized some transactions as being manually reviewed or labeled by workers in India. Ars Technica reported claims that human video review had been used, citing reporting from The Information (Ars Technica). Amazon disputed the idea that workers continuously watched shoppers, saying people were used for exceptions, system problems and training or labeling. Axios reported Amazon’s response and its continued third-party expansion (Axios).
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Both statements can be true at different levels. Human exception handling and data labeling are common in machine-learning operations; they do not prove that people perform most ordinary transactions. Conversely, the existence of automation does not prove that human intervention is rare. Amazon has not publicly quantified the scale, frequency or operational importance of review in the cited materials. Retail buyers should ask for exception rates, review rates, service-level targets and audit logs.
Why large grocery stores exposed the limits
In April 2024, Amazon said it was removing Just Walk Out from most full-size U.S. Amazon Fresh grocery stores in favor of Dash Carts. Its explanation was customer fit: shoppers making larger trips wanted a visible running total and more direct control of the basket. Dash Cart uses computer vision and sensor fusion while allowing customers to interact with the cart (Amazon’s explanation).
Independent GeekWire testing found that Dash Cart supplied more feedback but introduced its own friction, including scale warnings and demanding interactions (GeekWire comparison). The lesson is not that checkout-free technology works or fails universally; store format changes the economics and the error surface.
| Store environment | Likely fit | Why |
|---|---|---|
| Stadium concession | Strong | Speed matters and the assortment is limited. |
| Airport convenience | Strong | Customers are time-sensitive and baskets are small. |
| Small convenience store | Potentially strong | Controlled layouts and grab-and-go missions reduce ambiguity. |
| University or hospital shop | Potentially strong | Frequent quick purchases can justify queue reduction. |
| Full weekly grocery trip | More difficult | Many items, substitutions, bags, produce and bulky goods create more edge cases. |
| Large general merchandise store | Difficult | Assortment and customer behavior are more varied. |
Amazon’s 2026 position
Amazon’s retreat from its own grocery estate should not be described as a shutdown of Just Walk Out. In 2024, it reduced the technology’s role in many full-size Fresh stores. In an update dated January 27, 2026, Amazon said it was closing Amazon Go and Amazon Fresh physical stores and converting various locations to Whole Foods Market stores (Amazon’s January 27, 2026 update).
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchBest Value
- 【Core Parameters】★AI Perf: 117/157 TOPS★GPU: 1024-core N-VI-DIA Ampere architecture GPU with 32 Tensor Cores★CPU: 8-core Arm Cortex-A78AE v8.2 64-bit CPU 2MB L2 + 4MB L3★Memory: 16GB 128-bit LPDDR5 | 102.4GB/s★Storage: Supports external NVMe 【Note: This kit does not include a SSD and pre-installed system. User need to provide your own NVMe M.2 SSD of at least 256GB and flash the operating system onto it yourself. 】
- 【Empowered by Large Al Model, Enhanced Human-Computer Interaction】Jetson Orin Super leverages three AI models and incorporates an AI voice interaction module. This multimodal visual system matches the scene being described, enabling environmental awareness and AI visual gameplay. Combined with a large-scale voice module and camera, it enables speech-to-text, semantic analysis, natural conversation, and real-time video analysis, enabling advanced embodied AI applications.
- 【Revolutionize the Industry】Jetson Orin NX modules deliver unmatched performance and efficiency for small, low-power robotics and autonomous machines, making them ideal for drones, handheld devices, and more. The module can be easily used in advanced applications in manufacturing, logistics, retail, agriculture, medical and life sciences, and comes in a highly compact and energy-efficient package.
- 【Revolutionizing AI with Unmatched Performance】The Jetson Orin NX system module adopts the Ampere architecture GPU, a new generation of deep learning and vision accelerators, high-speed I/O, and fast memory bandwidth to support multiple AI application processes. Granular structured sparsity to improve the operating throughput of Tensor Core, and can use larger and more complex AI model development solutions in natural language understanding, 3D perception and multi-sensor fusion.
- 【Tutorial materials provided】The JETSON system based on Ubuntu 22.04 provides a complete desktop Linux environment with accelerated graphics, supporting NVIDI-ACUDA 12.6, TensorRT 10.7.0, cuDNN 9.6.0, OpenCV 4.10.0, etc. The performance on AI LLM, VLM and visual Transformer is significantly improved compared with the previous generation.
At the same time, Just Walk Out remains a commercial offering for third-party stadiums, airports, campuses, hospitals, convenience stores and other small-format locations (Just Walk Out). The strategic interpretation is narrower and more useful than “Amazon killed it”: Amazon is applying a complex perception-and-payment platform where high throughput and small baskets can justify the infrastructure.
What retailers should evaluate before buying
Customer mission and basket
- Measure average basket size and the share of one- or two-item purchases.
- Model how bags, strollers, shared carts and family accounts affect attribution.
- Test whether customers value leaving immediately more than seeing an itemized running total.
Products and layout
- Packaged, standardized products are generally easier than produce, variable-weight goods, apparel and lookalike packages.
- Document how the system handles misplaced stock, restocking while customers are present and frequent planogram changes.
- For RFID deployments, price tags, readers, installation and process changes separately.
Reliability and exceptions
- Ask what continues locally during a network outage and how receipts are reconciled afterward.
- Require false-positive, false-negative, dispute and manual-review metrics for a comparable deployment.
- Specify workflows for failed card authorization, refunds, age-restricted goods, tailgating and customers using the wrong exit.
- Require audit logs linking product events to receipts and a defined customer-support response time.
Privacy and governance
- Review video retention, access controls, payment handling and account-linking practices.
- Clarify whether Amazon One is optional and provide a non-biometric path where appropriate.
- Explain to customers what data is collected, why it is needed and how disputes are handled.
Total cost
There is no public standard list price in the cited Just Walk Out materials. Treat it as an enterprise, quote-based purchase. Total cost can include cameras, sensors, edge hardware, networking, RFID tags and readers, installation, software, integration, maintenance, support, store redesign and ongoing exception handling. Checkout-labor savings alone are not a complete business case.
Commercial alternatives and their trade-offs
| Approach | Customer interaction | Infrastructure burden | Best fit |
|---|---|---|---|
| Just Walk Out | Authenticate, take items, leave | Dense cameras and sensors, edge/cloud software, integrations | Small baskets and high-throughput venues |
| Dash Cart | Manage items and view a running total | Smart carts, computer vision and cart integration | Grocery and larger baskets |
| RFID checkout-free | Usually minimal at checkout | Tags, readers and merchandise-process changes | Apparel, footwear and tagged merchandise |
| Scan-and-go or self-checkout | Customer scans or confirms items | Lower automation complexity, greater customer participation | Retailers prioritizing control, transparency or lower deployment risk |
AWS identifies scan-and-go, RFID, IoT and Just Walk Out as distinct smart-store approaches (AWS smart-store overview). The technically least ambitious option can be financially superior if it delivers adequate queue reduction with fewer cameras, sensors and exception disputes.
What Amazon’s published numbers do—and do not—prove
Amazon reports theft losses below 1% in compliant stores from June 2024 through May 2025. That is an Amazon-published, condition-specific figure, not an independently audited industry benchmark (Amazon’s computer-vision article). Amazon also publishes selected customer case studies reporting revenue or transaction growth; those results belong to the named deployments and should not be generalized.
Amazon has not published a complete error taxonomy, universal false-positive or false-negative rate, average dispute rate, per-store exception rate, standard implementation cost or exact model latency. Those omissions matter because the commercial value depends on how often staff must repair the system and how costly each repair is.
The bottom line
Just Walk Out’s important development is not that a transformer magically made checkout autonomous. It is that Amazon is applying multimodal, physical-world AI through a hybrid edge/cloud architecture to infer actions, identity and payment events in a messy environment. The technology can be compelling where baskets are small, layouts are controlled and every second of queue time has economic value. Large grocery trips expose different needs—visibility, substitutions, bagging and customer control—which is why Amazon shifted many of its own stores toward Dash Cart while continuing to commercialize Just Walk Out for selected third-party formats.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
