ever wonder how amazon gets your packages to you so quickly it's done with a symphony of machine learning automation and our amazing associates join us as we see that in action in an amazon fulfillment center welcome to an amazon robotics sortable fulfillment center [Music] this fc is one of more than 50 worldwide where robots work alongside our amazing employees let's go inside at amazon we're customer obsessed we're also obsessed with the safety of our employees in april of 2020 amazon began enhancing its temperature screening process by implementing thermal camera systems throughout multiple business lines in
north america now that our temperature is not detected as high we're free to continue the tour [Music] today we're going to follow the path of products from where they're received stowed then picked packed and put on a truck for distribution the inbound process starts with forecasting and ordering before anything arrives to the fc [Music] powered by aws amazon's forecasting engine is used for over 400 million products daily let's go to the receiving area as you'll see we put a lot of energy into our logistics it's a symphony of people aws software and other high-tech components
the result is a coordination between our great employees and the finely honed computer systems we've evolved for more than 20 years our employees are the heart and soul of our operation some are military veterans some are active members of the community some are employees that work here with their loved ones they're all amazing trailers full of vendor and small and medium business seller items arrive at these bay doors by appointment employees unload the trailers and begin the process of staging items located at their receiving workstations this fc carries millions of items of inventory and using
the service known as amazon aurora we manage our inventory transactions and other relational database needs at this station you can see how our inbound associate receives a product and how those products are stowed in our inventory amazon robotic drive units are activated when an associate logs into their workstation to perform their task storage pods arrive facing the correct side to stow items into inventory every time an associate touches a product they scan a unique barcode for that item which we call the amazon standard identification number or asin that's stored in our inventory system so we
know what the quantity and location of that item is at any given time to manage our inventory history we use amazon neptune a fast reliable fully managed graph database service from aws now you might expect that we'd organize this in a way where similar items are stowed together for example all electronics would be stored in the same section however at amazon we've found that it's more efficient to stow products using a randomized method that's why you'll see a variety of items in each inventory bin once that process has been completed a physical and digital match
is created in our systems so just a few seconds after an item is stowed you can order it on amazon.com looking at a station next door we see a great example of computer vision helping with automation when an associate scans an item you'll see magenta lights over some of the bins in the pod this lets the associate know not to place that item in those locations as artificial intelligence driven logic has determined that the bin is full that placing an item there may negatively impact the pods overall weight distribution or that a visually similar item
is there and may cause confusion later when an associate needs to pick the item the associate can then place the item in any of the other bins but computer vision doesn't stop there it will also detect what bin the associate reached into and knows where that item is placed as one final step it will take a picture of the pod and use image recognition to count the number of items in the bin making sure the count matches what is recorded in our tracking systems when the machine learning model has low confidence it sends images to
people to classify and train more ground truth data for our machine learning model which operates in amazon sagemaker and amazon sagemaker ground truth from aws this allows us to perform ongoing inventory reconciliation avoiding the scheduled inventory counts that are often done at least once a year at other organizations now what's up with those robots we see moving the pods around there are thousands of robot drive units spread throughout each floor helping associates fulfill customer orders daily amazon robotics operate in dedicated fenced-off areas within the fc which account for about 65 of the facility's total square
footage the drive units travel to various locations in rfc to pick up mobile shelves of product and deliver them to employees who are working at ergonomic workstations to navigate they read 2d barcodes on the floor and locations are tracked in real time using our own created robotic operating software our amazon robotics team has developed more than 100 services to support operations and uses aws extensively in the amazon fulfillment centers only specially trained individuals wearing custom safety equipment are allowed to enter this area but on our virtual tour we're allowed to walk through this restricted area
each sortable fc has over one million unique items helping us with our goal to provide earth's largest selection you may be thinking okay robots are great but i just ordered an item how's it picked to answer that let's go to a pick station here we see a similar layout as our stow station once you hit the buy button on amazon.com our systems will cue the robots to find the inventory pod with the item you ordered once that item has been identified the robot will bring it to our associate a screen at the associate station will
show them a picture and quantity of the item they need to pick the same visual bin inspection system is used as during stow except this time instead of lighting up bins where items should not be placed the system lights up in a white light where that item is the associate identifies and grabs the item placing it in the yellow toad signified by the green light and pushes the button acknowledging that the item was placed there when the system detects that the tote is full the light flashes and the associate pushes the full tote back to
a series of conveyors and replaces it with a new one that was automatically brought to their station at this stage associates are picking items for different customers in batches the two types of picking processes we focus on are single item orders and multiple item orders the single item orders contain only one item per customer order we fill up a yellow tote full of single item orders weighing up to 25 pounds the single item order yellow totes will then be routed to the singles pack lines for packaging the multiple item orders contain more than one item
per order if your multiple item order of six items are all in this building then it's possible that your order is being picked up by up to six different people picking items from six different robots which all need to be brought together for your single shipment your items will be placed in any available yellow tote and routed to our multiple item order assembly stations it's at that point that we will reap in your individual items into one customer order then once all of your items are together it's packing time to help improve efficiency and also
reduce environmental impact of the packaging the size of boxer envelope is automatically selected by a machine learning algorithm even the correct length of tape is determined by machine learning and automatically dispensed if needed protective packaging materials are added a barcode is placed on the package it's sealed up and away we go that barcode will be important for the next stage that we call slab but what if i order two items and those items are in two different fcs do i get one box or two the answer is it depends our first priority is to get
you the item at the date and time that we promised when you placed the order the second priority is to do that at the lowest cost possible and with the lowest environmental impact so we can continue to pass on lower prices to you with those two things in mind we use machine learning to figure out if it's optimal to ship to you from each fc in two different packages or ship an item from one fc to the other combine them there and ship to you in a single package you're now seeing our slam line in
process slam is an acronym which stands for scan label apply and manifest the technology here scans that barcoded sticker placed on the package in pack and immediately knows to print the correct shipping label on the package this is the first time your name and address are placed on the package so how do we know which carrier to use that's also determined by machine learning in the space of a few seconds an algorithm is run to determine the optimal delivery carrier to get you the package on time and at the lowest cost possible there are also
quality control measures in place here following this the box is weighed and the weight is checked against the item's known weight if the weight is off the order will be routed to a special location from there an associate will open the package and take the applicable action we're now entering the final stages after passing all of those quality control measures the package is routed onto another series of conveyors to a location where the label is scanned the conveyors then automatically send each package down the correct shoot where it will be packaged on a semi-trailer for
shipment to another fc or a sortation center for final delivery to you there are approximately 17.5 miles of conveyor belts in this fc in order to run reliably every single day they're monitored via aws iot services aws iot green grass helps here by allowing devices to act locally on the data they generate while still taking advantage of the cloud events in particular from the machinery are sent to lambda functions in aws the trigger step functions these workflows then coordinate rme operations to maintain the equipment perform inspections replace faulty motors and belts and other types of
activities we even collaborate with some of our automation suppliers using aws ml models to predict equipment failures and address issues before they arise [Music] thank you for visiting our amazon sortable robotics fulfillment center we hope you enjoyed our virtual tour
