Amazon Prime Air has been in the news recently, on the heels of many other news stories about drones. Amazon touts that the system will be able to deliver packages in 30 minutes or less and that the system will become as normal as mail trucks on the road. Are these realistic claims, or is this just a publicity stunt?

While you may think Amazon’s system will be the first of its kind, the autonomous delivery of goods by drones has been underway since 2011 by the US Navy with its Cargo Unmanned Aircraft System (UAS). Although the payload these two systems carry is vastly different (thousands of pounds vs. five pounds), examining the Navy’s solution helps answer questions posed about Amazon’s proposed solution. While we wait to see if Amazon Prime Air will one day fly, let’s analyze how Amazon might approach the significant operational challenges they face in coordinated path planning, package delivery, and safety.

Coordinating Flight Operations

How can a  drone safely navigate from a fulfillment center to a customer’s house?  Sensing and avoiding terrain, obstacles, and other aircraft is a challenge, and one that the Federal Aviation Administration (FAA) is especially concerned about. A starting point is to use procedural separation, which relies on aircraft using separate routes to avoid each other. An analogous example is driving on the right side of the road so that you don’t hit oncoming traffic. Aircraft can separate through differing routes, altitudes, and flight times, and this goes a long way towards achieving safety, but can be operationally inefficient. Boundaries need to be determined ahead of time with built-in buffers to account for uncertainty.

Image: Flight Global / Aurora Flight Sciences

Image: Flight Global / Aurora Flight Sciences

In its initial operation, the Navy’s Cargo UAS system likely relies heavily on procedural separation, and this will be used by Prime Air too. However, it is not enough, as the environment can change, rendering well thought out procedures obsolete. An additional measure is to use a sensor to detect obstructions and fly around them. Onboard radar, while effective, is too large for small drones. Cameras, laser rangefinders, and even microphones have been tried for detecting obstructions with varying success, but these technologies are relatively new and will need further development before they are operationally mature.

Delivering the Package

Finding a relatively flat, obstacle-free landing spot for Prime Air drones is another challenge that the system will face. The Navy solves this by having a trained operator in the field directing the drone through its final maneuver. However, the next generation of the system (called “autonomous aerial cargo/utility system” or AACUS) will only require very minimally trained operators and will rely on onboard sensors to find suitable landing spots. While AACUS is designed to land in open fields, Prime Air will have to navigate more difficult urban or suburban environments with many houses, trees, and powerlines standing in the way. In dense urban areas, landing at ground level may in fact be impossible, and rooftop landings could provide a better alternative.


Addressing Drone Safety

Finally, the concerns that seem to be most discussed are safety and (possibly more importantly) the public’s perception of safety. The three aspects of safety that must be considered are the physical integrity of the aircraft, the autonomous system’s handling of abnormal situations, and the consequences of an accident. More importantly, the system must be designed from the outset with a safety mindset.

The two helicopters being used to test the Navy’s next-generation AACUS system are both derived from commercially available manned helicopters, addressing the first safety aspect. Amazon claims that their “vehicles will be built with multiple redundancies and designed to commercial aviation standards”. Such standards may exist in the future, but there are none yet that address quadrotor drones, as Amazon plans to use. Second, redundancies don’t always solve safety issues, as was seen with US Airways 1549, which simultaneously lost both engines and landed in the Hudson River. These are not insurmountable obstacles, but not trivial ones either.

Image: NY Daily News

Image: NY Daily News

The US Airways 1549 example segues nicely into a discussion of how abnormal situations are handled. Autonomous systems are ideal for performing mundane, repeated tasks, but humans are flexible decision makers who can handle complex, abnormal situations. The newest production cars (e.g. 2015 Mercedes C-Class) can maintain your speed in traffic, brake in advance of collisions, and keep you in your lane, but still rely on the driver in more abnormal conditions. Google’s latest self-driving car concept removes the steering wheel and pedals, leaving the operation of those controls to a computer, thereby moving the human to a higher level of decision making.

Similarly, drones have a human operator somewhere in the decision loop, but the difficulty lies in deciding how authority is divided between the drone and the human. With Prime Air, both safety and operating economics will drive this decision, as it will need to achieve a high drone to operator ratio for efficiency. A plausible scenario is that initial operations will be very carefully watched over, with the goal of moving to high-level supervisory control where drones have the logic to handle vehicle failures, but humans provide system-level control. This might include setting procedural separation rules mentioned earlier and modifying the tasks assigned to each vehicle. Academic research, such as at the Humans and Automation Lab (formerly at MIT, now at Duke University), is moving towards this paradigm.

On the last safety aspect, the consequences of an accident, the biggest concern is the environment over which the system will operate. In contrast to the remote areas over which the Navy operates, where an aircraft can crash without causing harm, Prime Air operates over dense suburban and urban environments.  Prime Air aircraft will need to be safe enough that the FAA and the public are reasonably assured that one will not come crashing down, of course leading back to Amazon’s statement that it will design its drones to commercial aviation standards.

The Future for Prime Air

The above analysis shows the challenges that Prime Air will face and hopefully gives the reader a better understanding of some of the issues surrounding such a system. However, the question remains whether such a system is worth the investment. The Navy’s Cargo UAS system was designed with the intent of taking troops and trucks off resupply roads vulnerable to ambush and improvised explosive devices (IEDs). Prime Air doesn’t have the same concerns, and is instead focused on speedy delivery. Over the range that Prime Air operates, could a system based on autonomous cars or trucks be more economical and versatile, and nearly as fast?