Ever wondered why Amazon
delivers your package to you so quickly? This is done by machine learning, automation and our amazing colleagues. Come with us, we will
see this in the Amazon fulfillment center. Welcome to the Amazon Robotic Sortable Operations Center. This FC is one of more than 50 FCs in the world Robots are here
to work with our outstanding employees. Let’s take a look inside. At Amazon, we are customer-centric. We are also very concerned about employee safety. In April 2020, Amazon adopted the implementation of thermal imaging systems in multiple business lines in North America. Begin to strengthen its body temperature screening process. Since our body temperature is not too high , we can continue the tour. Today, we will follow the path of products from receiving, storing, picking, packaging to loading on trucks for delivery before anything arrives at FC. The site process starts from forecasting and ordering. With the support of AWS, Amazon’s forecasting engine is used for more than 400 million products every day. Let’s go to the receiving area. As you can see, we have invested a lot of energy in logistics.
The symphony of , AWS, software and other high-tech components is the result of coordination between our outstanding employees and the well-honed computer system that we have developed for more than 20 years.
Our employees are the core and soul of our operations. Some are veterans. Some are active members of the community and some are employees working here with their relatives. They are amazing. Trailers filled with goods from suppliers and small and medium-sized business
sellers arrive at these hatches as scheduled. Employees unload the goods on the trailer
and start working there. The process of receiving goods temporarily stored on the workstation. This FC carries millions of pieces of inventory and uses
a service called Amazon Aurora to manage our
inventory transactions and other relational database needs. In this workbench, you can see how our inbound employees do Receiving products and how these products
are stored in our inventory. When employees log in to their workstations to perform tasks, the Amazon robot drive unit is activated. When the storage compartment arrives, it faces the right side to store the products in the inventory every time the employee touches the product.
They will scan the unique barcode of the item. We call it the Amazon Standard Identification Code or ASIN. It is stored in our inventory system
so we know the quantity and location of the item at any given time. To manage our inventory history,
we use Amazon. Neptune is
a fast, reliable, and fully managed graph database service provided by AWS. Now, you may want us to organize similar products together. For example, all electronic devices will be stored in the same interval. However, in Amazon , We found that it’s more effective to store products using random methods. That’s why you will
see various products in each inventory box. Once the process is completed , physical and digital matches will be created in our system. Therefore, store the products for a few seconds After the clock you can be on
Ordered on com. Look at a workbench next door. We saw a good example of computer vision
helping automation. When an employee scans an item, you will see a magenta light on some boxes in the storage compartment.
This makes the employee Knowing that the product cannot be placed in these locations because the artificial intelligence-driven logic determines that the box is full. Placing the product there may have a negative impact on the overall weight distribution of the storage compartment or there are visually similar products. When the employee needs to select the product later When the time, it may cause confusion. Then, the employee can put the product in any other box, but computer vision does not stop there. It will also detect the box used by the employee and know where the product is placed as the last step. It will take a picture of the storage compartment. And use image recognition to count the number of items in the box to ensure that the count matches the number recorded in our tracking system. When the confidence of the machine learning model is low, it will send the image to people in
order to classify our machine learning model and Train more Ground Truth data.
The model runs in Amazon SageMaker and Amazon SageMaker
Ground Truth on AWS.
This allows us to perform continuous inventory checks to avoid the scheduled inventory counts that other organizations tend to do at least once a year. Now, we see
those mobile storage How is the robot in the cabin? Thousands of robot drive devices on each floor help employees complete customer orders every day. Amazon robots
operate in dedicated fenced areas in FC. These areas account for about 65% of the total area of the facility. Drive devices go to various locations in our FC to pick up mobile shelves.
For navigation, they read the two-dimensional barcode on the floor and use our own robot operation software to track the position in real time. Our Amazon robotics team
developed 100 A number of services to support operations and extensive use of AWS in Amazon operation centers. Only specially trained personnel wearing customized safety equipment can enter the area. However, in our virtual tour we can pass through this restricted area. Every sortable FC
is available. There are more than one million unique products that help us achieve our goal of providing the most choices on the planet. You might think: "Well, robots are great, but I just ordered one product.
How did it get picked?" To answer this question, let’s go to a picking station. We see
a layout similar to our loading station. After you click the buy button on Amazon.com,
our system will queue the robot to find the product you ordered.
Once the storage compartment of the product is determined, the robot will bring it to our colleagues . The screen of the staff station will display the picture and quantity of the product they need to select. The same visual box inspection system used during loading. Except this time, the system will not Light up the box where the product should not be placed, but light up the white light where the product is located. The employee identifies and grabs the product and puts it in the yellow handbag indicated by the green light , and then presses the button to confirm that the product has been placed in it.
There the indicator light flashes when the system detects that the handbag is full , and the employees push the filled handbag back to a series of conveyor belts and replace it with a new handbag that is automatically delivered to their site. At this stage, the employees are divided into different batches Of customers picking products . The two picking processes we pay attention to are single-product orders and multi-product orders. In single-product orders, each customer order contains only one product. We filled a yellow handbag with a single product weighing 25 pounds. The single-product order yellow handbag will then be sent to the
Singles Pack production line for packaging. In a multi-product order, each order contains multiple products. If your 6-piece single-product order is in this building, then your order may be Up to 6 different people
pick up goods from 6 different robots and these robots all need to
be brought together for your single shipment.
Your goods will be placed in any available yellow tote bag and shipped to us At the time of the multi-product order assembly station , we will repack your personal products into a customer order. Then, once all your products are put together,
they can be packaged in order to help improve efficiency and reduce packaging Impact on the environment. The machine learning algorithm will automatically select the size of the box or envelope. Even the correct tape length is determined by machine learning and automatically assigned. If necessary, you can add protective packaging materials , paste a barcode on the package, and seal it. Then we can go. This barcode is
very important for the next stage we call SLAM. But what if I order two items and they are in two different FCs? Do I want one box or two? The answer is "it depends".
Our first priority is that when you place an order, we will provide you with the goods on the promised date and time. The second priority is to have the lowest possible cost and minimal impact on the environment. Way to do this so we can continue to pass on the lower price to you. Considering these two points, we use machine learning to determine whether to ship to you in two different packages from each FC
or to ship the goods from one FC Go to another where they are combined
and shipped to you in a single package. You can now see
our SLAM production line is in progress. SLAM is the acronym
for Scan, Label, Apply and Manifest . The technical scan here is placed in the package. It’s the first time your name and address appear on the package so how do we know which carrier to use? This is also determined by machine learning to run an algorithm within a few seconds to determine the best delivery carrier, so as to provide you with the package on time at the lowest possible cost. There are also quality control measures. After this, the box is weighed and Check according to the known weight of the goods.
If the weight is less, the
order will be routed to a special location. From there, an employee will open the package and take the appropriate action. We are now in the final stage. After passing all these quality control measures, the package will be It is transferred to another series of conveyor belts to reach the position where the labels are scanned, and then the conveyor belt will automatically send each package into the correct chute, and then pack it with a semi-trailer and then transport it to another FC or sorting center and
finally deliver it to you. About 17.5 miles of the conveyor belt in the center, in order to operate reliably, they are monitored daily through the AWS IoT service. AWS IoT Greengrass
allows devices to operate locally on the data they generate while still using the cloud, thereby providing help events, especially from machines The events are sent to AWS to
trigger the Lambda function of Step Functions. Then, these workflows coordinate RME operations to maintain equipment, perform inspections and replace faulty motors and belts, and other types of activities.
We even cooperate with some of our automation suppliers. AWS ML model to predict equipment failures and solve them before they occur. Thank you for visiting our
Amazon sortable robot operation center. We hope you enjoy our virtual tour.