Bezos's Drones

Jeff Bezos has announced that Amazon.com may begin delivering products with unmanned drones. They're estimating 30-minute delivery times using a little computer with eight helicopter-style propellers, just in time for Christmas 2015.

Is this really going to happen? If it did, what would it look like? I found this idea fascinating, so here's my uneducated guess at how the logistics will play out.

Ground rules
We can guess a few things about this plan already, if it does ultimately come to fruition.

High-cost option
Amazon.com, like all online retailers, already offers multiple levels of shipping options in which greater cost offsets faster delivery. If you want your package overnighted, you'll pay something like $10-$20, but if you're willing to wait your turn in the back of a United States Postal Service truck, you might not have to pay for shipping at all. (Of course, there's no such thing as a free lunch, so in reality even the "free" shipping options are included in the base price of what you buy, but that's not strictly relevant.)

Extrapolating from the current shipping option cost structure, a 30-minute shipping option would fit in at the very highest end. Anyone asking to receive a purchase faster than overnight should have to pay more than the overnight shipping cost&mdash;otherwise there would be no reason to pick anything other than drone delivery. Expect $20-$50, maybe (at least until the whole operation is fully rolled out and optimized, at which point it might start to displace some of the other options). This means that the infrastructure can be given a lot of leeway with respect to its costs, at least initially. Expensive support systems are not to be ruled out at first; we have $50 with which to work per round trip, give or take. Since the users of this service are likely to have an urgent need for it, there is also some flexibility on the upside of the pricing; $50 could become $80 or $100 if the system isn't initially cost effective.

Limits to cargo
Any drone will have an upper limit to how much weight it can lift, and probably an even lower limit to how much weight it can transport safely and efficiently (there's no point in taking off if the thing can only fly at 5% of its top speed and has to stop to recharge every mile). Expect to see the "drone delivery" option wink in and out as you add and remove more items to your order.

Similarly, the risk of the drone being jostled by wind or obstacles is simply too high to make it worth trying to ship fragile goods. Amazon.com will happily fly a few books to you, but don't expect the drone option to be available for the fine china you're gifting your mother in law.

(Apparently the official figure is 5 pounds.)

Geography
It won't be physically possible to have a drone within 30 minutes' flight time of every single household in the United States, let alone the entire world. To be cost effective, the program would have to target areas with the highest population densities, fanning out into further flung areas as it scales up, or with disclaimers about taking more than 30 minutes. You may have to input your shipping address before the web site gives you the option to ship by drone.

The system will probably also require good wireless coverage, to ensure Amazon.com can maintain contact with its fleet at all times. Expect that to play a role in availability as well.

(Apparently the official figure is 10 miles.)

Weather
A rainy day might just be drab for you, but for Prime Air it means the drones will come back from their missions dripping wet. Either the infrastructure needs to be weather-resistant, or there may be limits on when and where it's available (i.e., never in Seattle).

Low throughput
The temptation in designing a semi-automated shipping system is to squeeze out every sort of delay and inefficiency, but given the above natural limits, the number of packages being flown by drones should be a small percentage of Amazon.com's overall sales, possibly not enough to keep even the smallest pipeline busy continuously. It looks like primarily a high-status entertainment option, or potentially a component of some particularly time-sensitive just-in-time business models.

Logistics
Very well, how will it all work? Only Amazon.com knows for sure, but I enjoy speculating.

Smart upstream routing
When you enter your address and choose drone delivery, Amazon.com's servers will calculate the best drone location to use based on location, capacity, and availability of products. Your order will then be routed directly to this location for processing. This all seems well within Amazon.com's legendary IT capabilities.

Warehouse rooms
The first time your order interacts with a human is when your shipping label prints out in the precise location of the product you ordered. I envision warehouses divided into smaller rooms organized by type of product, each monitored by one employee, who: At this point, your purchase is on its way to be transported, the warehouse employee is waiting for the next order, and it all took less than 2 minutes.
 * 1) Tears the label off the printer
 * 2) Reads the codes for box size and what goes in it
 * 3) Grabs the needed box off one of the stacks near the printer
 * 4) Heads to the aisles of shelves
 * 5) Walks up to the location of the purchased product
 * 6) Scans its bar code for confirmation and tracking
 * 7) Puts it in the box
 * 8) Returns to the printer
 * 9) Affixes the shipping label and seals the box with tape
 * 10) Drops it on the conveyor belt to the drone bay

Drone bay
Your box rides the conveyor belt up to the roof of the warehouse where an airport baggage claim-style ramp dumps it into a bin. A drone bay employee: The drone bay employee's job is now done, and they are waiting for the next package. Again it took less than 2 minutes.
 * 1) Picks up the box
 * 2) Grabs a fresh drone battery from the charger near the conveyor belt drop-off
 * 3) Walks to an available drone at its docking station
 * 4) Inserts the fresh battery to allow the drone to boot up
 * 5) Attaches the package securely
 * 6) Uses the docking station's bar code scanner to scan the bar code on the shipping label

International pilot pool
Finally the fun part, the flying! But wait, I thought these were going to be unmanned drones?? Yes and no. I expect that most municipalities will be very uncomfortable with fully automated aircraft flying near their citizens, but most of that discomfort will evaporate if the automated part of the journey only takes place high overhead (1000 ft? I dunno, I'm not a pilot). So I think we'll see a situation where Prime Air flies on autopilot for most of each trip, with human pilots patching in just for the sensitive low-altitude parts at each end.

I picture a room similar to a call center where Prime Air's pilots sit in front of computers or game consoles wired into the drone network, handling these brief maneuvers repeatedly through the day. There can be one centralized pilot facility, or dozens distributed widely; their location and staffing levels can be adjusted as needed since it doesn't matter from where they remotely operate the drones.

Take-off
When the ground crew bar-codes a package, the attached drone broadcasts to the wireless network that it's ready to take off. This causes it to appear in the pilots' pending queue, presumably labeled as a take-off request. One of the pilots: The pilot then returns to the request queue and waits for the next task, and the drone flies toward its drop-off point on its own.
 * 1) Presses a button to patch him- or herself in to the ground camera at the relevant drone bay
 * 2) If the pilot sees a problem, he or she can alert the ground crew
 * 3) Otherwise, the pilot presses another button to accept the mission, which takes it out of the pending queue that other pilots can see
 * 4) The pilot uses a keyboard, joystick, mouse, gamepad, or other device to guide the drone out of the drone bay and into the sky
 * 5) Once the drone reaches cruising altitude, a beep inaugurates the 10-second countdown to fully automatic mode
 * 6) If everything looks OK, the pilot can wait for the countdown or patch out early

Delivery
As the drone approaches its destination, it again signals for human assistance. A pilot (who has been handling other requests continuously during the drone's flight): The pilot again goes back to the task queue, and the drone flies back to base on autopilot.
 * 1) Goes through the same procedure as before to patch in, check the drone's situation, and accept the mission, this time using the drone's own on-board downward-facing camera only
 * 2) Identifies the delivery destination using on-screen indicators based on the bar coded address and map data
 * 3) Begins lowering the drone to a safe drop-off altitude
 * 4) Tells the drone to lower the package on its retractable tether (this allows the drop-off to occur gently while the drone is still too high to be a hazard to humans)
 * 5) Slows the drone's descent as the package approaches the ground or porch
 * 6) Achieves a gentle touch-down of the package
 * 7) Disconnects and retracts the tether; the package is now at its destination!
 * 8) Commands the drone to ascend
 * 9) Again the drone initiates a countdown to automatic mode once it reaches cruising altitude
 * 10) The pilot may patch out anytime after cruising altitude is reached

Homecoming
For the third and final time during its outing, the drone signals for pilot take-over as it approaches the drone bay. The drone is now ready for a new shipment, and both the pilot and the ground crew member are waiting for their next tasks.
 * 1) A pilot patches in as before, guides the drone down to a docking station, and patches out again
 * 2) A ground crew member ensures the drone is secured and removes its battery
 * 3) The battery is placed in the recharging station

Subtleties and assumptions
Prime Air alludes ominously to the Pentagon and CIA's military drone programs. It seems likely that Amazon.com will be receiving some form of assistance, either in the form of direct subsidies to undertake the program (worth spending from the Pentagon's perspective to increase public acceptance of drones), used hardware, technical expertise, or at the very least a market where drones can be bought cheaper than they otherwise would be after years of building businesses that serve the military.

The drone bays could be located at Amazon.com's current warehouse facilities, or at new ones dedicated solely to drone delivery. The advantage of the latter is that drone-shippable products could be concentrated in a smaller space; this could also be achieved in general-purpose facilities by sorting the inventory, but that might cut the efficiency of non-drone shipments.

Each location would have to designate a staff member responsible for retrieving errant drones and their cargo in case of a crash (and presumably completing the delivery by hand). This would not need to be any more complicated than a car with a GPS and a sensor to track the drone. This responsibility could be included as an auxiliary task in the ground crew role.

I have assumed blithely that the FAA's regulations of domestic drone flights will be updated to allow Prime Air because that legal question does not interest me.

Navigation
More densely populated areas also tend to have taller buildings, which could add some complexity to the autopilot's flight logic. Either the cruising altitude would have to be sufficiently high, or the drones would have to know to navigate around obstacles.

Drone-mounted cameras will add significant weight and cost and so must be minimized, but at least one seems necessary for precise drop-offs. But what about avoiding dynamic obstacles (e.g., other Prime Air drones)? Some kind of additional proximity sensor may be needed.

Synchronization and load balancing
My human intervention steps (warehouse, drone bay, take-off, drop-off) seem to require at least 5-10 minutes, so the realistic flight range is more like 20-25 minutes rather than 30.

If the pilots are all busy, any drones ready for assistance can hover at altitude while they wait in the queue. Since each flight task is relatively short, such waiting times can be minimized as needed by staffing more pilots. The upstream database should be able to detect a sudden surge in drone delivered orders and alert managers to the need to call in "on call" pilots. The biggest risk is that customers would not like to see a drone hovering stationary above their residences for several minutes.

Regarding drop-offs, the current convention of delivering a package and waiting for customers to find it presumably would continue. An active alert system may be needed, though, since packages will of necessity be deposited in less convenient locations than would be used by human delivery staff, and the very use of Prime Air suggests customer urgency anyway. Perhaps an email could be sent as the pilot patches out after drop-off.

Power
The batteries don't need to be removed if their capacity can handle several consecutive flights between charges. However, it's somewhat advantageous to make the landing and take-off procedures completely uniform by skipping the decision-making step of checking how depleted the battery is, so it would make sense to choose the cheapest (and lightest!) batteries that could reliably handle one round trip plus idle padding (see below). Similarly, the batteries could be left inside each drone and recharged by the docking station, but this introduces an extra decision point where the ground crew has to check whether the drone is ready and potentially wait; it would be more consistent to always remove the batteries, and quicker to simply pre-charge a few extras for a quick swap-out.

Drone-mounted solar panels could reduce the need to recharge the batteries, but they would also add weight and so may or may not be worthwhile.

(Apparently the state of the art is 30 minutes.)

Further automation
Some of the human steps could be automated further; for example, a robotic arm could retrieve pre-packaged items from storage shelves and place them into a label-applicator device, or automatic loading of drones could be attempted. However, I don't think the demand for Prime Air will be great enough to justify such expensive technology, and the error rate at such tasks would probably be far lower for humans.

(Apparently the drones can pick up packages themselves off of conveyor belts.)

Anticipatory shipping changes everything!
Amazon.com has filed a patent for "anticipatory shipping", in which their infrastructure would move products around in anticipation of them being bought in particular places. The Gizmodo article makes it sound ridiculous, as if they would try to tailor it exactly to your individual tastes and send you things you hadn't bought yet or had almost bought, but in fact it makes sense and can mesh beautifully with drones!

Consider the launch of a new niche product; maybe it's a new Xbox or Android phone. We know beforehand that lots of them will be bought and shipped in a big surge on day one. Market research can tell you how demand for that product is distributed across various demographic groups, and census data can tell you how those demographic groups are distributed geographically. Combine the two, mix in Amazon.com's historical purchase records (for guessing what proportion of the affected population will follow through with a purchase), and it should be possible to estimate how many of a given product will be purchased in a given city, and even which neighborhoods will have the highest density of purchases. The biggest unknown is exactly which individuals in that city will make a purchase (and even that can be anticipated when pre-orders are available).


 * 1) A fleet of drones takes off before dawn, each carrying a package containing one of that day's anticipatorily-shipped products, with no set destination yet
 * 2) The drones hover at altitude over the areas with the greatest probability of purchases, waiting
 * 3) You log in to Amazon.com's web servers
 * 4) The server looks up what you're likely to purchase from your customer profile, geolocates your primary shipping address, and broadcasts a notification to nearby drone piloting centers
 * 5) If your likely purchases match up with that day's anticipatory shipping products, an available drone is commanded to fly towards your house
 * 6) You click to view details about the product in question
 * 7) The drone accelerates toward your residence
 * 8) You add the product to your online shopping cart
 * 9) The drone begins its descent
 * 10) You hear a faint buzzing. You look out the window. It's there, hovering.
 * 11) You check out and pay
 * 12) The drone resumes its descent
 * 13) Your credit card payment is validated
 * 14) The drone lands on your porch, deposits its payload, and returns home for the next package that Amazon.com anticipates someone in your area will buy

The upside for Amazon.com is increased customer satisfaction and loyalty, as it becomes clear that buying such a product from them is a sure way to get it quickly.

Wal*Mart has its "warehouse on wheels", but Amazon.com could do them one better with a "warehouse in the air".